net: remove unneeded tags from dnsRR structs

DNS packing and unpacking uses hand-coded struct walking functions
rather than reflection, so these tags are unneeded and just contribute
to their runtime reflect metadata size.

Change-Id: I2db09d5159912bcbc3b482cbf23a50fa8fa807fa
Reviewed-on: https://go-review.googlesource.com/22594
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

net: remove internal support for obsolete DNS record types

There are no real world use cases for HINFO, MINFO, MB, MG, or MR
records, and package net's exposed APIs don't provide any way to
access them even if there were. If a use ever does show up, we can
revive them. In the mean time, this is just effectively-dead code that
sticks around because of rr_mk.

Change-Id: I6c188b5ee32f3b3a04588b79a0ee9c2e3e725ccc
Reviewed-on: https://go-review.googlesource.com/22593
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

net/http, net/http/httptrace: new package for tracing HTTP client requests

Updates #12580

Change-Id: I9f9578148ef2b48dffede1007317032d39f6af55
Reviewed-on: https://go-review.googlesource.com/22191
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Tom Bergan <tombergan@google.com>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

os/exec: fix variable shadow, don't leak goroutine

Goroutine leak checking is still too tedious, so untested.

See #6705 which is my fault for forgetting to mail out.

Change-Id: I899fb311c9d4229ff1dbd3f54fe307805e17efee
Reviewed-on: https://go-review.googlesource.com/22581
Reviewed-by: Ahmed W. <oneofone@gmail.com>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

cmd/compile: use delta encoding for filenames in export data position info

This reduces the export data size significantly (15%-25%) for some packages,
especially where the paths are very long or if there are many files involved.
Slight (2%) reduction on average, with virtually no increases in export data
size.

Selected export data sizes for packages with |delta %| > 3%:

                     package   before    after   delta     %

        cmd/asm/internal/arch   11647    11088    -559   -4%
   cmd/compile/internal/amd64     838      600    -238  -27%
     cmd/compile/internal/arm    7323     6793    -530   -6%
   cmd/compile/internal/arm64   19948    18971    -977   -4%
     cmd/compile/internal/big    9043     8548    -495   -4%
  cmd/compile/internal/mips64     645      482    -163  -24%
   cmd/compile/internal/ppc64     695      497    -198  -27%
   cmd/compile/internal/s390x     553      433    -120  -21%
     cmd/compile/internal/x86     744      555    -189  -24%
                     cmd/dist     145      121     -24  -16%
         cmd/internal/objfile   17359    16474    -885   -4%
   cmd/internal/pprof/symbolz    8346     7941    -405   -4%
      cmd/link/internal/amd64   11178    10604    -574   -4%
        cmd/link/internal/arm     204      171     -33  -15%
      cmd/link/internal/arm64     210      175     -35  -16%
     cmd/link/internal/mips64     213      177     -36  -16%
      cmd/link/internal/ppc64     211      176     -35  -16%
      cmd/link/internal/s390x     210      175     -35  -16%
        cmd/link/internal/x86     203      170     -33  -15%
                    cmd/trace     782      744     -38   -4%
                 compress/lzw     402      383     -19   -4%
                   crypto/aes     311      262     -49  -15%
                crypto/cipher    1138      959    -179  -15%
                   crypto/des     315      288     -27   -8%
              crypto/elliptic    6063     5746    -317   -4%
                   crypto/rc4     317      295     -22   -6%
                crypto/sha256     348      312     -36   -9%
                crypto/sha512     487      451     -36   -6%
                       go/doc    3871     3649    -222   -5%
    go/internal/gccgoimporter    2063     1949    -114   -5%
       go/internal/gcimporter    3253     3096    -157   -4%
                         math    4343     3572    -771  -17%
                   math/cmplx    1580     1274    -306  -18%
                    math/rand     982      926     -56   -5%
        net/internal/socktest    2159     2049    -110   -4%
                      os/exec    7928     7492    -436   -4%
                    os/signal     237      208     -29  -11%
                      os/user     717      682     -35   -4%
      runtime/internal/atomic     728      693     -35   -4%
         runtime/internal/sys    2287     2107    -180   -7%
                         sync    1306     1214     -92   -6%

                 all packages 1509255  1465507  -43748   -2%

Change-Id: I98a11521b552166b7f47f2039a29f106748bf5d4
Reviewed-on: https://go-review.googlesource.com/22580
Reviewed-by: Alan Donovan <adonovan@google.com>

cmd/compile: remove unused Bputname function

Change-Id: Icecbf9bae8c39670d1ceef62dd94b36e90b27b04
Reviewed-on: https://go-review.googlesource.com/22570
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

cmd/compile: allow 64-bit multiplication with immediates on s390x

MGHI (16-bit signed immediate) is now used where possible for both
MULLW and MULLD. MGHI is 2-bytes shorter than MSGFI.

Change-Id: I5d0648934f28b3403b1126913fd703d8f62b9e9f
Reviewed-on: https://go-review.googlesource.com/22398
Run-TryBot: Michael Munday <munday@ca.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Bill O'Farrell <billotosyr@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

net: ensure dnsConfig search list is rooted

Avoids some extra work and string concatenation at query time.

benchmark                                      old allocs     new allocs     delta
BenchmarkGoLookupIP-32                         154            150            -2.60%
BenchmarkGoLookupIPNoSuchHost-32               446            442            -0.90%
BenchmarkGoLookupIPWithBrokenNameServer-32     564            568            +0.71%

benchmark                                      old bytes     new bytes     delta
BenchmarkGoLookupIP-32                         10824         10704         -1.11%
BenchmarkGoLookupIPNoSuchHost-32               43140         42992         -0.34%
BenchmarkGoLookupIPWithBrokenNameServer-32     46616         46680         +0.14%

BenchmarkGoLookupIPWithBrokenNameServer's regression appears to be
because it's actually only performing 1 LookupIP call, so the extra
work done parsing the DNS config file doesn't amortize as well as for
BenchmarkGoLookupIP or BenchmarkGoLOokupIPNoSuchHost, which perform
2000+ LookupIP calls per run.

Update #15473.

Change-Id: I98c8072f2f39e2f2ccd6c55e9e9bd309f5ad68f8
Reviewed-on: https://go-review.googlesource.com/22571
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

net: append ":53" to DNS servers when reading resolv.conf

Avoids generating some redundant garbage from re-concatenating the
same string for every DNS query.

benchmark                                      old allocs     new allocs     delta
BenchmarkGoLookupIP-32                         156            154            -1.28%
BenchmarkGoLookupIPNoSuchHost-32               456            446            -2.19%
BenchmarkGoLookupIPWithBrokenNameServer-32     577            564            -2.25%

benchmark                                      old bytes     new bytes     delta
BenchmarkGoLookupIP-32                         10873         10824         -0.45%
BenchmarkGoLookupIPNoSuchHost-32               43303         43140         -0.38%
BenchmarkGoLookupIPWithBrokenNameServer-32     46824         46616         -0.44%

Update #15473.

Change-Id: I3b0173dfedf31bd08eaea1069968b416850864a1
Reviewed-on: https://go-review.googlesource.com/22556
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

os/exec: add Cmd.RunContext and Cmd.WaitContext

Updates #14660

Change-Id: Ifa5c97ba327ad7ceea0a9a252e3dbd9d079dae54
Reviewed-on: https://go-review.googlesource.com/22529
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

crypto/tls: allow renegotiation to be handled by a client.

This change adds Config.Renegotiation which controls whether a TLS
client will accept renegotiation requests from a server. This is used,
for example, by some web servers that wish to “add” a client certificate
to an HTTPS connection.

This is disabled by default because it significantly complicates the
state machine.

Originally, handshakeMutex was taken before locking either Conn.in or
Conn.out. However, if renegotiation is permitted then a handshake may
be triggered during a Read() call. If Conn.in were unlocked before
taking handshakeMutex then a concurrent Read() call could see an
intermediate state and trigger an error. Thus handshakeMutex is now
locked after Conn.in and the handshake functions assume that Conn.in is
locked for the duration of the handshake.

Additionally, handshakeMutex used to protect Conn.out also. With the
possibility of renegotiation that's no longer viable and so
writeRecordLocked has been split off.

Fixes #5742.

Change-Id: I935914db1f185d507ff39bba8274c148d756a1c8
Reviewed-on: https://go-review.googlesource.com/22475
Run-TryBot: Adam Langley <agl@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>

cmd/compile: reorg copyelim to avoid O(n^2) problem

Make sure we don't do O(n^2) work to eliminate a chain
of n copies.

benchmark                     old ns/op       new ns/op     delta
BenchmarkCopyElim1-8          1418            1406          -0.85%
BenchmarkCopyElim10-8         5289            5162          -2.40%
BenchmarkCopyElim100-8        52618           41684         -20.78%
BenchmarkCopyElim1000-8       2473878         424339        -82.85%
BenchmarkCopyElim10000-8      269373954       6367971       -97.64%
BenchmarkCopyElim100000-8     31272781165     104357244     -99.67%

Change-Id: I680f906f70f2ee1a8615cb1046bc510c77d59284
Reviewed-on: https://go-review.googlesource.com/22535
Reviewed-by: Alexandru Moșoi <alexandru@mosoi.ro>

cmd/compile: fix.gc.Type.cmp for map.notBucket cmp map.Bucket

Comparison of certain map types could fail to be antisymmetric.
This corrects that.

Change-Id: I88c6256053ce29950ced4ba4d538e241ee8591fe
Reviewed-on: https://go-review.googlesource.com/22552
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: jcd . <jcd@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>

net: change type of dnsConfig.timeout from int to time.Duration

Instead of keeping the desired number of seconds and converting to
time.Duration for every query, convert to time.Duration when
building the config.

Updates #15473

Change-Id: Ib24c050b593b3109011e359f4ed837a3fb45dc65
Reviewed-on: https://go-review.googlesource.com/22548
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

cmd/vet: improve checking unkeyed fields in composite literals

- Simplified the code.

- Removed types for slice aliases from composite literals' whitelist, since they
are properly handled by vet.

Fixes #15408
Updates #9171
Updates #11041

Change-Id: Ia1806c9eb3f327c09d2e28da4ffdb233b5a159b0
Reviewed-on: https://go-review.googlesource.com/22318
Run-TryBot: Rob Pike <r@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rob Pike <r@golang.org>

cmd/compile: have all or no parameter named in exported signatures

Binary export format only.

Make sure we don't accidentally export an unnamed parameter
in signatures which expect all named parameters; otherwise
we crash during import. Appears to happen for _ (blank)
parameter names, as observed in method signatures such as
the one at: x/tools/godoc/analysis/analysis.go:76.

Fixes #15470.

TBR=mdempsky

Change-Id: I1b1184bf08c4c09d8a46946539c4b8c341acdb84
Reviewed-on: https://go-review.googlesource.com/22543
Reviewed-by: Robert Griesemer <gri@golang.org>

cmd/compile: use correct (field/method) node for position info

Position info for fields and methods was based on the wrong node
in the new export format, leading to position info for empty
file names and 0 line numbers. Use correct node now.

Due to compact delta encoding, there is no difference in export
format size. In fact, because encoding of "no line changed" is
uncommon and thus a bit more expensive, in many cases the data
is now slightly shorter.

Stats for export data size (pachage, before, after, delta%):

                                         archive/tar     5128     5025  -1%
                                         archive/zip     7672     7515  -1%
                                               bufio     3486     3377  -2%
                                               bytes     4879     4821   0%
                                       cmd/addr2line       66       66   0%
                                             cmd/api    14033    13970   0%
                                             cmd/asm       60       60   0%
                               cmd/asm/internal/arch    11659    11647   0%
                                cmd/asm/internal/asm    13858    13766   0%
                              cmd/asm/internal/flags      306      300  -1%
                                cmd/asm/internal/lex    15684    15623   0%
                                             cmd/cgo    15383    15298   0%
                                         cmd/compile       63       63   0%
                          cmd/compile/internal/amd64      838      838   0%
                            cmd/compile/internal/arm     7333     7323   0%
                          cmd/compile/internal/arm64    19953    19951   0%
                            cmd/compile/internal/big     8943     9043   1%
                             cmd/compile/internal/gc    57465    56801   0%
                         cmd/compile/internal/mips64      645      645   0%
                          cmd/compile/internal/ppc64      695      695   0%
                          cmd/compile/internal/s390x      553      553   0%
                            cmd/compile/internal/ssa    34883    34559   0%
                            cmd/compile/internal/x86      744      744   0%
                                           cmd/cover     4961     4892   0%
                                            cmd/dist      145      145   0%
                                             cmd/doc     8891     8853   0%
                                         cmd/expdump       67       67   0%
                                             cmd/fix      422      406  -3%
                                              cmd/go     9951     9747  -1%
                                           cmd/gofmt       66       66   0%
                                    cmd/internal/bio     6378     6340   0%
                                 cmd/internal/gcprog      684      644  -5%
                                  cmd/internal/goobj     1276     1193  -6%
                                    cmd/internal/obj    12908    12551  -2%
                                cmd/internal/obj/arm    10074    10053   0%
                              cmd/internal/obj/arm64    17723    17699   0%
                               cmd/internal/obj/mips    12573    12530   0%
                              cmd/internal/obj/ppc64    15352    15330   0%
                              cmd/internal/obj/s390x    18785    18769   0%
                                cmd/internal/obj/x86    23586    23551   0%
                                cmd/internal/objfile    17148    17359   1%
                         cmd/internal/pprof/commands     1948     1930   0%
                           cmd/internal/pprof/driver    11123    11095   0%
                            cmd/internal/pprof/fetch     8931     8907   0%
                           cmd/internal/pprof/plugin    15335    15221   0%
                          cmd/internal/pprof/profile     8493     8370   0%
                           cmd/internal/pprof/report     9273     9214   0%
                              cmd/internal/pprof/svg     1589     1589   0%
                       cmd/internal/pprof/symbolizer     8737     8727   0%
                          cmd/internal/pprof/symbolz     8277     8346   1%
                         cmd/internal/pprof/tempfile     4319     4317   0%
                                    cmd/internal/sys      622      603  -2%
  cmd/internal/unvendor/golang.org/x/arch/arm/armasm    79231    79148   0%
  cmd/internal/unvendor/golang.org/x/arch/x86/x86asm    11761    11726   0%
                                            cmd/link       60       60   0%
                             cmd/link/internal/amd64    11190    11178   0%
                               cmd/link/internal/arm      204      204   0%
                             cmd/link/internal/arm64      210      210   0%
                                cmd/link/internal/ld    60670    59758  -1%
                            cmd/link/internal/mips64      213      213   0%
                             cmd/link/internal/ppc64      211      211   0%
                             cmd/link/internal/s390x      210      210   0%
                               cmd/link/internal/x86      203      203   0%
                                              cmd/nm       57       57   0%
                                         cmd/objdump       64       64   0%
                                            cmd/pack     4968     4908   0%
                                           cmd/pprof       63       63   0%
                                           cmd/trace      828      782  -5%
                                             cmd/vet    13485    13503   0%
                          cmd/vet/internal/whitelist      109      109   0%
                                            cmd/yacc     1315     1269  -2%
                                      compress/bzip2     2561     2506  -1%
                                      compress/flate     4906     4748  -2%
                                       compress/gzip     7788     7717   0%
                                        compress/lzw      406      402   0%
                                       compress/zlib     4739     4712   0%
                                      container/heap      265      257  -2%
                                      container/list     1506     1450  -3%
                                      container/ring      556      536  -3%
                                             context     3552     3527   0%
                                              crypto      864      834  -2%
                                          crypto/aes      313      311   0%
                                       crypto/cipher     1139     1138   0%
                                          crypto/des      317      315   0%
                                          crypto/dsa     5326     5304   0%
                                        crypto/ecdsa     6383     6364   0%
                                     crypto/elliptic     5983     6063   1%
                                         crypto/hmac      258      256   0%
                                          crypto/md5      722      700  -2%
                                         crypto/rand     4996     4993   0%
                                          crypto/rc4      327      317  -2%
                                          crypto/rsa     6763     6722   0%
                                         crypto/sha1      767      744  -2%
                                       crypto/sha256      348      348   0%
                                       crypto/sha512      487      487   0%
                                       crypto/subtle      620      620   0%
                                          crypto/tls    24344    24083   0%
                                         crypto/x509    17473    17524   0%
                                    crypto/x509/pkix     9682     9596   0%
                                        database/sql     8099     7831  -2%
                                 database/sql/driver     1556     1500  -3%
                                         debug/dwarf     9358     9010  -3%
                                           debug/elf    28226    27882   0%
                                         debug/gosym     2472     2333  -5%
                                         debug/macho     9032     8830  -1%
                                            debug/pe     8561     8328  -2%
                                      debug/plan9obj     1347     1295  -3%
                                            encoding      275      261  -4%
                                    encoding/ascii85      775      738  -4%
                                       encoding/asn1     1280     1246  -2%
                                     encoding/base32     1207     1146  -4%
                                     encoding/base64     1471     1407  -3%
                                     encoding/binary     2430     2386  -1%
                                        encoding/csv     4347     4280  -1%
                                        encoding/gob    13488    13387   0%
                                        encoding/hex      665      646  -2%
                                       encoding/json    11763    11592   0%
                                        encoding/pem      283      273  -3%
                                        encoding/xml    13804    13631   0%
                                              errors      166      162  -1%
                                              expvar     1193     1139  -4%
                                                flag     6896     6964   1%
                                                 fmt     1247     1213  -2%
                                              go/ast    15797    15473  -1%
                                            go/build     6497     6336  -1%
                                         go/constant     1846     1820   0%
                                              go/doc     3942     3871  -1%
                                           go/format     1854     1850   0%
                                         go/importer     1702     1695   0%
                           go/internal/gccgoimporter     2084     2063   0%
                              go/internal/gcimporter     3236     3253   1%
                                           go/parser     7377     7371   0%
                                          go/printer     2480     2469   0%
                                          go/scanner     3806     3733  -1%
                                            go/token     3579     3523  -1%
                                            go/types    26514    26117   0%
                                                hash      323      295  -8%
                                        hash/adler32      568      554  -1%
                                          hash/crc32      843      825  -1%
                                          hash/crc64      758      739  -2%
                                            hash/fnv     1583     1530  -2%
                                                html      113      113   0%
                                       html/template    16957    16937   0%
                                               image    11470    11045  -3%
                                         image/color     2566     2503  -1%
                                 image/color/palette      165      163   0%
                                          image/draw     2543     2522   0%
                                           image/gif     3467     3439   0%
                            image/internal/imageutil     3481     3479   0%
                                          image/jpeg     2725     2717   0%
                                           image/png     2702     2689   0%
                                   index/suffixarray     5802     5777   0%
                                       internal/race      274      274   0%
                               internal/singleflight      756      718  -4%
                     internal/syscall/windows/sysdll      162      162   0%
                                    internal/testenv     5288     5276   0%
                                      internal/trace     1853     1768  -4%
                                                  io     3425     3349  -1%
                                           io/ioutil     4768     4756   0%
                                                 log     4173     4224   1%
                                          log/syslog     5049     4996   0%
                                                math     4343     4343   0%
                                            math/big     8779     8817   0%
                                          math/cmplx     1580     1580   0%
                                           math/rand      944      982   4%
                                                mime     2313     2298   0%
                                      mime/multipart     5021     4922  -1%
                                mime/quotedprintable     2049     2008  -1%
                                                 net    19332    19090   0%
                                            net/http    50404    49542  -1%
                                        net/http/cgi    22533    22637   0%
                                  net/http/cookiejar     5488     5431   0%
                                       net/http/fcgi    20557    20512   0%
                                   net/http/httptest    30350    30255   0%
                                   net/http/httputil    24045    23964   0%
                                   net/http/internal     2579     2550   0%
                                      net/http/pprof    20307    20258   0%
                               net/internal/socktest     2227     2159  -2%
                                            net/mail     5086     5054   0%
                                             net/rpc    28365    28208   0%
                                     net/rpc/jsonrpc    12805    12722   0%
                                            net/smtp    19975    19887   0%
                                       net/textproto     4558     4466  -1%
                                             net/url     1391     1326  -4%
                                                  os    10372    10195  -1%
                                             os/exec     7814     7928   1%
                                           os/signal      239      237   0%
                                             os/user      735      717  -1%
                                                path      391      391   0%
                                       path/filepath     4136     4136   0%
                                             reflect     6258     5893  -5%
                                              regexp     5808     5623  -2%
                                       regexp/syntax     3118     3077   0%
                                             runtime    11685    10912  -6%
                                         runtime/cgo       18       18   0%
                                       runtime/debug     3320     3304   0%
                             runtime/internal/atomic      728      728   0%
                                runtime/internal/sys     2287     2287   0%
                                       runtime/pprof      611      587  -3%
                                        runtime/race       19       19   0%
                                       runtime/trace      145      143   0%
                                                sort     1229     1206  -1%
                                             strconv     1752     1744   0%
                                             strings     3809     3775   0%
                                                sync     1331     1306  -1%
                                         sync/atomic     1135     1130   0%
                                             syscall    46280    45722   0%
                                             testing     7558     7284  -3%
                                      testing/iotest     1122     1071  -4%
                                       testing/quick     5656     5609   0%
                                        text/scanner     3367     3312  -1%
                                      text/tabwriter     2810     2755  -1%
                                       text/template    15613    15595   0%
                                 text/template/parse     9499     9040  -4%
                                                time     5515     5395  -1%
                                             unicode     4357     4344   0%
                                       unicode/utf16      583      583   0%
                                        unicode/utf8      970      970   0%
                 vendor/golang.org/x/net/http2/hpack     4105     4012  -1%
                                             average  1524284  1509610   0%

Change-Id: Ibe1ce098c7c575965389c1cad368c62c2cea256a
Reviewed-on: https://go-review.googlesource.com/22536
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Alan Donovan <adonovan@google.com>

net: fix misrecongnization of IPv6 zone on Windows

Fixes #15463.

Change-Id: Ic85886861c650ffcb71240d847941534152b92bc
Reviewed-on: https://go-review.googlesource.com/22540
Run-TryBot: Mikio Hara <mikioh.mikioh@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

net/http: skip TestTransportRemovesDeadIdleConnections on Plan 9

Updates #15464.

Change-Id: If3221034bb10751c6fcf1fbeba401a879c18079f
Reviewed-on: https://go-review.googlesource.com/22513
Run-TryBot: David du Colombier <0intro@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

cmd/compile/internal/gc: unexport {J,S,F,H,B,V}conv

Updates #15462

Unexport Jconv, Sconv, Fconv, Hconv, Bconv, and VConv as they are
not referenced outside internal/gc.

Econv was only called by EType.String, so merge it into that method.

Change-Id: Iad9b06078eb513b85a03a43cd9eb9366477643d1
Reviewed-on: https://go-review.googlesource.com/22531
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Dave Cheney <dave@cheney.net>
TryBot-Result: Gobot Gobot <gobot@golang.org>

cmd/compile/internal/gc: remove all uses of oconv(op, FmtSharp)

Updates #15462

Replace all use of oconv(op, FmtSharp) with fmt.Printf("%#v", op).
This removes all the callers of oconv.

Change-Id: Ic3bf22495147f8497c8bada01d681428e2405b0e
Reviewed-on: https://go-review.googlesource.com/22530
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

net: clarify DialContext's use of its provided context

Fixes #15325

Change-Id: I60137ecf27e236e97734b1730ce29ab23e9fe07f
Reviewed-on: https://go-review.googlesource.com/22509
Reviewed-by: Ian Lance Taylor <iant@golang.org>

runtime: fix SetCgoTraceback doc indentation

It wasn't rendering as HTML nicely.

Change-Id: I5408ec22932a05e85c210c0faa434bd19dce5650
Reviewed-on: https://go-review.googlesource.com/22532
Reviewed-by: Ian Lance Taylor <iant@golang.org>

crypto/md5: add s390x assembly implementation

Adapted from md5block_amd64.s.

name                 old speed      new speed      delta
Hash8Bytes           14.0MB/s ± 1%  39.9MB/s ± 0%  +185.52%   (p=0.000 n=9+10)
Hash1K                176MB/s ± 1%   661MB/s ± 1%  +274.44%  (p=0.000 n=10+10)
Hash8K                196MB/s ± 0%   742MB/s ± 1%  +278.35%   (p=0.000 n=10+9)
Hash8BytesUnaligned  14.2MB/s ± 2%  39.8MB/s ± 0%  +180.06%  (p=0.000 n=10+10)
Hash1KUnaligned       177MB/s ± 1%   651MB/s ± 0%  +267.38%  (p=0.000 n=10+10)
Hash8KUnaligned       197MB/s ± 1%   731MB/s ± 1%  +271.73%  (p=0.000 n=10+10)

Change-Id: I45ece98ee10f30fcd192b9c3d743ba61c248f36a
Reviewed-on: https://go-review.googlesource.com/22505
Reviewed-by: Bill O'Farrell <billotosyr@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

cmd/compile: de-dup the gclocals symbols in compiler too

These symbols are de-duplicated in the linker but the compiler generates quite
many duplicates too: 2425 of 13769 total symbols for runtime.a for example.
De-duplicating them in the compiler saves the linker a bit of work.

Fixes #14983

Change-Id: I5f18e5f9743563c795aad8f0a22d17a7ed147711
Reviewed-on: https://go-review.googlesource.com/22293
Reviewed-by: David Crawshaw <crawshaw@golang.org>

cmd/compile/internal/gc: remove oconv(op, 0) calls

Updates #15462

Automatic refactor with sed -e.

Replace all oconv(op, 0) to string conversion with the raw op value
which fmt's %v verb can print directly.

The remaining oconv(op, FmtSharp) will be replaced with op.GoString and
%#v in the next CL.

Change-Id: I5e2f7ee0bd35caa65c6dd6cb1a866b5e4519e641
Reviewed-on: https://go-review.googlesource.com/22499
Run-TryBot: Dave Cheney <dave@cheney.net>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

net: search domain from hostname if no search directives

Fixes #14897

Change-Id: Iffe7462983a5623a37aa0dc6f74c8c70e10c3244
Reviewed-on: https://go-review.googlesource.com/21464
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>

syscall: fix uint64->int cast of control message header

Change-Id: I28980b307d10730b122a4f833809bc400d6aff24
Reviewed-on: https://go-review.googlesource.com/22525
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

misc/cgo/testcarchive: fix path of libgo.a for darwin/arm

After CL 22461, c-archive build on darwin/arm is by default compiled
with -shared, so update the install path.

Fix build.

Change-Id: Ie93dbd226ed416b834da0234210f4b98bc0e3606
Reviewed-on: https://go-review.googlesource.com/22507
Reviewed-by: David Crawshaw <crawshaw@golang.org>
Run-TryBot: David Crawshaw <crawshaw@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

runtime: don't rescan globals

Currently the runtime rescans globals during mark 2 and mark
termination. This costs as much as 500µs/MB in STW time, which is
enough to surpass the 10ms STW limit with only 20MB of globals.

It's also basically unnecessary. The compiler already generates write
barriers for global -> heap pointer updates and the regular write
barrier doesn't check whether the slot is a global or in the heap.
Some less common write barriers do cause problems.
heapBitsBulkBarrier, which is used by typedmemmove and related
functions, currently depends on having access to the pointer bitmap
and as a result ignores writes to globals. Likewise, the
reflect-related write barriers reflect_typedmemmovepartial and
callwritebarrier ignore non-heap destinations; though it appears they
can never be called with global pointers anyway.

This commit makes heapBitsBulkBarrier issue write barriers for writes
to global pointers using the data and BSS pointer bitmaps, removes the
inheap checks from the reflection write barriers, and eliminates the
rescans during mark 2 and mark termination. It also adds a test that
writes to globals have write barriers.

Programs with large data+BSS segments (with pointers) aren't common,
but for programs that do have large data+BSS segments, this
significantly reduces pause time:

name \ 95%ile-time/markTerm              old         new  delta
LargeBSS/bss:1GB/gomaxprocs:4  148200µs ± 6%  302µs ±52%  -99.80% (p=0.008 n=5+5)

This very slightly improves the go1 benchmarks:

name                      old time/op    new time/op    delta
BinaryTree17-12              2.62s ± 3%     2.62s ± 4%    ~     (p=0.904 n=20+20)
Fannkuch11-12                2.15s ± 1%     2.13s ± 0%  -1.29%  (p=0.000 n=18+20)
FmtFprintfEmpty-12          48.3ns ± 2%    47.6ns ± 1%  -1.52%  (p=0.000 n=20+16)
FmtFprintfString-12          152ns ± 0%     152ns ± 1%    ~     (p=0.725 n=18+18)
FmtFprintfInt-12             150ns ± 1%     149ns ± 1%  -1.14%  (p=0.000 n=19+20)
FmtFprintfIntInt-12          250ns ± 0%     244ns ± 1%  -2.12%  (p=0.000 n=20+18)
FmtFprintfPrefixedInt-12     219ns ± 1%     217ns ± 1%  -1.20%  (p=0.000 n=19+20)
FmtFprintfFloat-12           280ns ± 0%     281ns ± 1%  +0.47%  (p=0.000 n=19+19)
FmtManyArgs-12               928ns ± 0%     923ns ± 1%  -0.53%  (p=0.000 n=19+18)
GobDecode-12                7.21ms ± 1%    7.24ms ± 2%    ~     (p=0.091 n=19+19)
GobEncode-12                6.07ms ± 1%    6.05ms ± 1%  -0.36%  (p=0.002 n=20+17)
Gzip-12                      265ms ± 1%     265ms ± 1%    ~     (p=0.496 n=20+19)
Gunzip-12                   39.6ms ± 1%    39.3ms ± 1%  -0.85%  (p=0.000 n=19+19)
HTTPClientServer-12         74.0µs ± 2%    73.8µs ± 1%    ~     (p=0.569 n=20+19)
JSONEncode-12               15.4ms ± 1%    15.3ms ± 1%  -0.25%  (p=0.049 n=17+17)
JSONDecode-12               53.7ms ± 2%    53.0ms ± 1%  -1.29%  (p=0.000 n=18+17)
Mandelbrot200-12            3.97ms ± 1%    3.97ms ± 0%    ~     (p=0.072 n=17+18)
GoParse-12                  3.35ms ± 2%    3.36ms ± 1%  +0.51%  (p=0.005 n=18+20)
RegexpMatchEasy0_32-12      72.7ns ± 2%    72.2ns ± 1%  -0.70%  (p=0.005 n=19+19)
RegexpMatchEasy0_1K-12       246ns ± 1%     245ns ± 0%  -0.60%  (p=0.000 n=18+16)
RegexpMatchEasy1_32-12      72.8ns ± 1%    72.5ns ± 1%  -0.37%  (p=0.011 n=18+18)
RegexpMatchEasy1_1K-12       380ns ± 1%     385ns ± 1%  +1.34%  (p=0.000 n=20+19)
RegexpMatchMedium_32-12      115ns ± 2%     115ns ± 1%  +0.44%  (p=0.047 n=20+20)
RegexpMatchMedium_1K-12     35.4µs ± 1%    35.5µs ± 1%    ~     (p=0.079 n=18+19)
RegexpMatchHard_32-12       1.83µs ± 0%    1.80µs ± 1%  -1.76%  (p=0.000 n=18+18)
RegexpMatchHard_1K-12       55.1µs ± 0%    54.3µs ± 1%  -1.42%  (p=0.000 n=18+19)
Revcomp-12                   386ms ± 1%     381ms ± 1%  -1.14%  (p=0.000 n=18+18)
Template-12                 61.5ms ± 2%    61.5ms ± 2%    ~     (p=0.647 n=19+20)
TimeParse-12                 338ns ± 0%     336ns ± 1%  -0.72%  (p=0.000 n=14+19)
TimeFormat-12                350ns ± 0%     357ns ± 0%  +2.05%  (p=0.000 n=19+18)
[Geo mean]                  55.3µs         55.0µs       -0.41%

Change-Id: I57e8720385a1b991aeebd111b6874354308e2a6b
Reviewed-on: https://go-review.googlesource.com/20829
Run-TryBot: Austin Clements <austin@google.com>
Reviewed-by: Rick Hudson <rlh@golang.org>

runtime: make {add,subtract}{b,1} nosplit

These are used at the bottom level of various GC operations that must
not be preempted. To be on the safe side, mark them all nosplit.

Change-Id: I8f7360e79c9852bd044df71413b8581ad764380c
Reviewed-on: https://go-review.googlesource.com/22504
Run-TryBot: Austin Clements <austin@google.com>
Reviewed-by: Rick Hudson <rlh@golang.org>

reflect: fix strings of SliceOf-created types

The new type was inheriting the tflagExtraStar from its prototype.

Fixes #15467

Change-Id: Ic22c2a55cee7580cb59228d52b97e1c0a1e60220
Reviewed-on: https://go-review.googlesource.com/22501
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>

reflect: unnamed interface types have no name

Fixes #15468

Change-Id: I8723171f87774a98d5e80e7832ebb96dd1fbea74
Reviewed-on: https://go-review.googlesource.com/22524
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: David Crawshaw <crawshaw@golang.org>

cmd/compile: enable const division for arm64

performance:
benchmark                   old ns/op     new ns/op     delta
BenchmarkDivconstI64-8      8.28          2.70          -67.39%
BenchmarkDivconstU64-8      8.28          4.69          -43.36%
BenchmarkDivconstI32-8      8.28          6.39          -22.83%
BenchmarkDivconstU32-8      8.28          4.43          -46.50%
BenchmarkDivconstI16-8      5.17          5.17          +0.00%
BenchmarkDivconstU16-8      5.33          5.34          +0.19%
BenchmarkDivconstI8-8       3.50          3.50          +0.00%
BenchmarkDivconstU8-8       3.51          3.50          -0.28%

Fixes #15382

Change-Id: Ibce7b28f0586d593b33c4d4ecc5d5e7e7c905d13
Reviewed-on: https://go-review.googlesource.com/22292
Reviewed-by: Michael Munday <munday@ca.ibm.com>
Reviewed-by: David Chase <drchase@google.com>

cmd/compile: switch to compact export format by default

builtin.go was auto-generated via go generate; all other
changes were manual.

The new format reduces the export data size by ~65% on average
for the std library packages (and there is still quite a bit of
room for improvement).

The average time to write export data is reduced by (at least)
62% as measured in one run over the std lib, it is likely more.

The average time to read import data is reduced by (at least)
37% as measured in one run over the std lib, it is likely more.
There is also room to improve this time.

The compiler transparently handles both packages using the old
and the new format.

Comparing the -S output of the go build for each package via
the cmp.bash script (added) shows identical assembly code for
all packages, but 6 files show file:line differences:

The following files have differences because they use cgo
and cgo uses different temp. directories for different builds.
Harmless.

src/crypto/x509
src/net
src/os/user
src/runtime/cgo

The following files have file:line differences that are not yet
fully explained; however the differences exist w/ and w/o new export
format (pre-existing condition). See issue #15453.

src/go/internal/gccgoimporter
src/go/internal/gcimporter

In summary, switching to the new export format produces the same
package files as before for all practical purposes.

How can you tell which one you have (if you care): Open a package
(.a) file in an editor. Textual export data starts with a $$ after
the header and is more or less legible; binary export data starts
with a $$B after the header and is mostly unreadable. A stand-alone
decoder (for debugging) is in the works.

In case of a problem, please first try reverting back to the old
textual format to determine if the cause is the new export format:

For a stand-alone compiler invocation:
- go tool compile -newexport=0 <files>

For a single package:
- go build -gcflags="-newexport=0" <pkg>

For make/all.bash:
- (export GO_GCFLAGS="-newexport=0"; sh make.bash)

Fixes #13241.

Change-Id: I2588cb463be80af22446bf80c225e92ab79878b8
Reviewed-on: https://go-review.googlesource.com/22123
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Robert Griesemer <gri@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>

regexp: add a harder regexp to the benchmarks

This regexp has many parallel alternations

Change-Id: I8044f460aa7d18f20cb0452e9470557b87facd6d
Reviewed-on: https://go-review.googlesource.com/22471
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

cmd/link: remove absolute address for c-archive on darwin/arm

Now it is possible to build a c-archive as PIC on darwin/arm (this is
now the default). Then the system linker can link the binary using
the archive as PIE.

Fixes #12896.

Change-Id: Iad84131572422190f5fa036e7d71910dc155f155
Reviewed-on: https://go-review.googlesource.com/22461
Reviewed-by: David Crawshaw <crawshaw@golang.org>

cmd/compile: don't write pos info for builtin packages

TestBuiltin will fail if run on Windows and builtin.go was generated
on a non-Windows machine (or vice versa) because path names have
different separators. Avoid problem altogether by not writing pos
info for builtin packages. It's not needed.

Affects -newexport only.

Change-Id: I8944f343452faebaea9a08b5fb62829bed77c148
Reviewed-on: https://go-review.googlesource.com/22498
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>

cmd/compile: don't use line numbers from ONAME and named OLITERALs

The line numbers of ONAMEs are the location of their
declaration, not their use.

The line numbers of named OLITERALs are also the location
of their declaration.

Ignore both of these.  Instead, we will inherit the line number from
the containing syntactic item.

Fixes #14742
Fixes #15430

Change-Id: Ie43b5b9f6321cbf8cead56e37ccc9364d0702f2f
Reviewed-on: https://go-review.googlesource.com/22479
Reviewed-by: Robert Griesemer <gri@golang.org>
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>

cmd/asm: fix SIMD register name on arm64

Current V-register range is V32~V63 on arm64. This patch changes it to
V0~V31.

fix #15465.

Change-Id: I90dab42dea46825ec5d7a8321ec4f6550735feb8
Reviewed-on: https://go-review.googlesource.com/22520
Reviewed-by: Aram Hăvărneanu <aram@mgk.ro>
Run-TryBot: Aram Hăvărneanu <aram@mgk.ro>
TryBot-Result: Gobot Gobot <gobot@golang.org>

runtime/race: improve TestNoRaceIOHttp test

TestNoRaceIOHttp does all kinds of bad things:
1. Binds to a fixed port, so concurrent tests fail.
2. Registers HTTP handler multiple times, so repeated tests fail.
3. Relies on sleep to wait for listen.

Fix all of that.

Change-Id: I1210b7797ef5e92465b37dc407246d92a2a24fe8
Reviewed-on: https://go-review.googlesource.com/19953
Run-TryBot: Dmitry Vyukov <dvyukov@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

image/color: optimize RGBToYCbCr

Apply optimizations used to speed up YCbCrToRGB from
https://go-review.googlesource.com/#/c/21910/
to RGBToYCbCr.

name             old time/op  new time/op  delta
RGBToYCbCr/0-2   6.81ns ± 0%  5.96ns ± 0%  -12.48%  (p=0.000 n=38+50)
RGBToYCbCr/Cb-2  7.68ns ± 0%  6.13ns ± 0%  -20.21%  (p=0.000 n=50+33)
RGBToYCbCr/Cr-2  6.84ns ± 0%  6.04ns ± 0%  -11.70%  (p=0.000 n=39+42)

Updates #15260

Change-Id: If3ea5393ae371a955ddf18ab226aae20b48f9692
Reviewed-on: https://go-review.googlesource.com/22411
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ralph Corderoy <ralph@inputplus.co.uk>

cmd/compile/internal: unexport gc.Oconv

Updates #15462

Semi automatic change with gofmt -r and hand fixups for callers outside
internal/gc.

All the uses of gc.Oconv outside cmd/compile/internal/gc were for the
Oconv(op, 0) form, which is already handled the Op.String method.

Replace the use of gc.Oconv(op, 0) with op itself, which will call
Op.String via the %v or %s verb. Unexport Oconv.

Change-Id: I84da2a2e4381b35f52efce427b2d6a3bccdf2526
Reviewed-on: https://go-review.googlesource.com/22496
Run-TryBot: Dave Cheney <dave@cheney.net>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>

cmd/compile: fix opnames

Change-Id: Ief4707747338912216a8509b1adbf655c8ffac56
Reviewed-on: https://go-review.googlesource.com/22495
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

net/http: remove idle transport connections from Transport when server closes

Previously the Transport would cache idle connections from the
Transport for later reuse, but if a peer server disconnected
(e.g. idle timeout), we would not proactively remove the *persistConn
from the Transport's idle list, leading to a waste of memory
(potentially forever).

Instead, when the persistConn's readLoop terminates, remote it from
the idle list, if present.

This also adds the beginning of accounting for the total number of
idle connections, which will be needed for Transport.MaxIdleConns
later.

Updates #15461

Change-Id: Iab091f180f8dd1ee0d78f34b9705d68743b5557b
Reviewed-on: https://go-review.googlesource.com/22492
Reviewed-by: Andrew Gerrand <adg@golang.org>

context: fix doc typo

Fixes #15449

Change-Id: I8d84d076a05c56694b48f7b84f572b1a6524f522
Reviewed-on: https://go-review.googlesource.com/22493
Reviewed-by: Andrew Gerrand <adg@golang.org>

cmd/go: add Package.StaleReason for debugging with go list

It comes up every few months that we can't understand why
the go command is rebuilding some package.
Add diagnostics so that the go command can explain itself
if asked.

For #2775, #3506, #12074.

Change-Id: I1c73b492589b49886bf31a8f9d05514adbd6ed70
Reviewed-on: https://go-review.googlesource.com/22432
Reviewed-by: Rob Pike <r@golang.org>

crypto/sha256: add s390x assembly implementation

Renames block to blockGeneric so that it can be called when the
assembly feature check fails. This means making block a var on
platforms without an assembly implementation (similar to the sha1
package).

Also adds a test to check that the fallback path works correctly
when the feature check fails.

name        old speed      new speed       delta
Hash8Bytes  6.42MB/s ± 1%  27.14MB/s ± 0%  +323.01%  (p=0.000 n=10+10)
Hash1K      53.9MB/s ± 0%  511.1MB/s ± 0%  +847.57%   (p=0.000 n=10+9)
Hash8K      57.1MB/s ± 1%  609.7MB/s ± 0%  +967.04%  (p=0.000 n=10+10)

Change-Id: If962b2a5c9160b3a0b76ccee53b2fd809468ed3d
Reviewed-on: https://go-review.googlesource.com/22460
Run-TryBot: Michael Munday <munday@ca.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Bill O'Farrell <billotosyr@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

runtime: make stack re-scan O(# dirty stacks)

Currently the stack re-scan during mark termination is O(# stacks)
because we enqueue a root marking job for every goroutine. It takes
~34ns to process this root marking job for a valid (clean) stack, so
at around 300k goroutines we exceed the 10ms pause goal. A non-trivial
portion of this time is spent simply taking the cache miss to check
the gcscanvalid flag, so simply optimizing the path that handles clean
stacks can only improve this so much.

Fix this by keeping an explicit list of goroutines with dirty stacks
that need to be rescanned. When a goroutine first transitions to
running after a stack scan and marks its stack dirty, it adds itself
to this list. We enqueue root marking jobs only for the goroutines in
this list, so this improves stack re-scanning asymptotically by
completely eliminating time spent on clean goroutines.

This reduces mark termination time for 500k idle goroutines from 15ms
to 238µs. Overall performance effect is negligible.

name \ 95%ile-time/markTerm     old           new         delta
IdleGs/gs:500000/gomaxprocs:12  15000µs ± 0%  238µs ± 5%  -98.41% (p=0.000 n=10+10)

name              old time/op  new time/op  delta
XBenchGarbage-12  2.30ms ± 3%  2.29ms ± 1%  -0.43%  (p=0.049 n=17+18)

name                      old time/op    new time/op    delta
BinaryTree17-12              2.57s ± 3%     2.59s ± 2%    ~     (p=0.141 n=19+20)
Fannkuch11-12                2.09s ± 0%     2.10s ± 1%  +0.53%  (p=0.000 n=19+19)
FmtFprintfEmpty-12          45.3ns ± 3%    45.2ns ± 2%    ~     (p=0.845 n=20+20)
FmtFprintfString-12          129ns ± 0%     127ns ± 0%  -1.55%  (p=0.000 n=16+16)
FmtFprintfInt-12             123ns ± 0%     119ns ± 1%  -3.24%  (p=0.000 n=19+19)
FmtFprintfIntInt-12          195ns ± 1%     189ns ± 1%  -3.11%  (p=0.000 n=17+17)
FmtFprintfPrefixedInt-12     193ns ± 1%     187ns ± 1%  -3.06%  (p=0.000 n=19+19)
FmtFprintfFloat-12           254ns ± 0%     255ns ± 1%  +0.35%  (p=0.001 n=14+17)
FmtManyArgs-12               781ns ± 0%     770ns ± 0%  -1.48%  (p=0.000 n=16+19)
GobDecode-12                7.00ms ± 1%    6.98ms ± 1%    ~     (p=0.563 n=19+19)
GobEncode-12                5.91ms ± 1%    5.92ms ± 0%    ~     (p=0.118 n=19+18)
Gzip-12                      219ms ± 1%     215ms ± 1%  -1.81%  (p=0.000 n=18+18)
Gunzip-12                   37.2ms ± 0%    37.4ms ± 0%  +0.45%  (p=0.000 n=17+19)
HTTPClientServer-12         76.9µs ± 3%    77.5µs ± 2%  +0.81%  (p=0.030 n=20+19)
JSONEncode-12               15.0ms ± 0%    14.8ms ± 1%  -0.88%  (p=0.001 n=15+19)
JSONDecode-12               50.6ms ± 0%    53.2ms ± 2%  +5.07%  (p=0.000 n=17+19)
Mandelbrot200-12            4.05ms ± 0%    4.05ms ± 1%    ~     (p=0.581 n=16+17)
GoParse-12                  3.34ms ± 1%    3.30ms ± 1%  -1.21%  (p=0.000 n=15+20)
RegexpMatchEasy0_32-12      69.6ns ± 1%    69.8ns ± 2%    ~     (p=0.566 n=19+19)
RegexpMatchEasy0_1K-12       238ns ± 1%     236ns ± 0%  -0.91%  (p=0.000 n=17+13)
RegexpMatchEasy1_32-12      69.8ns ± 1%    70.0ns ± 1%  +0.23%  (p=0.026 n=17+16)
RegexpMatchEasy1_1K-12       371ns ± 1%     363ns ± 1%  -2.07%  (p=0.000 n=19+19)
RegexpMatchMedium_32-12      107ns ± 2%     106ns ± 1%  -0.51%  (p=0.031 n=18+20)
RegexpMatchMedium_1K-12     33.0µs ± 0%    32.9µs ± 0%  -0.30%  (p=0.004 n=16+16)
RegexpMatchHard_32-12       1.70µs ± 0%    1.70µs ± 0%  +0.45%  (p=0.000 n=16+17)
RegexpMatchHard_1K-12       51.1µs ± 2%    51.4µs ± 1%  +0.53%  (p=0.000 n=17+19)
Revcomp-12                   378ms ± 1%     385ms ± 1%  +1.92%  (p=0.000 n=19+18)
Template-12                 64.3ms ± 2%    65.0ms ± 2%  +1.09%  (p=0.001 n=19+19)
TimeParse-12                 315ns ± 1%     317ns ± 2%    ~     (p=0.108 n=18+20)
TimeFormat-12                360ns ± 1%     337ns ± 0%  -6.30%  (p=0.000 n=18+13)
[Geo mean]                  51.8µs         51.6µs       -0.48%

Change-Id: Icf8994671476840e3998236e15407a505d4c760c
Reviewed-on: https://go-review.googlesource.com/20700
Reviewed-by: Rick Hudson <rlh@golang.org>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

runtime: don't clear gcscanvalid in casfrom_Gscanstatus

Currently we clear gcscanvalid in both casgstatus and
casfrom_Gscanstatus if the new status is _Grunning. This is very
important to do in casgstatus. However, this is potentially wrong in
casfrom_Gscanstatus because in this case the caller doesn't own gp and
hence the write is racy. Unlike the other _Gscan statuses, during
_Gscanrunning, the G is still running. This does not indicate that
it's transitioning into a running state. The scan simply hasn't
happened yet, so it's neither valid nor invalid.

Conveniently, this also means clearing gcscanvalid is unnecessary in
this case because the G was already in _Grunning, so we can simply
remove this code. What will happen instead is that the G will be
preempted to scan itself, that scan will set gcscanvalid to true, and
then the G will return to _Grunning via casgstatus, clearing
gcscanvalid.

This fix will become necessary shortly when we start keeping track of
the set of G's with dirty stacks, since it will no longer be
idempotent to simply set gcscanvalid to false.

Change-Id: I688c82e6fbf00d5dbbbff49efa66acb99ee86785
Reviewed-on: https://go-review.googlesource.com/20669
Reviewed-by: Rick Hudson <rlh@golang.org>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

runtime: fix typos in comment about gcscanvalid

Change-Id: Id4ad7ebf88a21eba2bc5714b96570ed5cfaed757
Reviewed-on: https://go-review.googlesource.com/22210
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

runtime: remove stack barriers during sweep

This adds a best-effort pass to remove stack barriers immediately
after the end of mark termination. This isn't necessary for the Go
runtime, but should help external tools that perform stack walks but
aren't aware of Go's stack barriers such as GDB, perf, and VTune.
(Though clearly they'll still have trouble unwinding stacks during
mark.)

Change-Id: I66600fae1f03ee36b5459d2b00dcc376269af18e
Reviewed-on: https://go-review.googlesource.com/20668
Reviewed-by: Rick Hudson <rlh@golang.org>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

runtime: remove stack barriers during concurrent mark

Currently we remove stack barriers during STW mark termination, which
has a non-trivial per-goroutine cost and means that we have to touch
even clean stacks during mark termination. However, there's no problem
with leaving them in during the sweep phase. They just have to be out
by the time we install new stack barriers immediately prior to
scanning the stack such as during the mark phase of the next GC cycle
or during mark termination in a STW GC.

Hence, move the gcRemoveStackBarriers from STW mark termination to
just before we install new stack barriers during concurrent mark. This
removes the cost from STW. Furthermore, this combined with concurrent
stack shrinking means that the mark termination scan of a clean stack
is a complete no-op, which will make it possible to skip clean stacks
entirely during mark termination.

This has the downside that it will mess up anything outside of Go that
tries to walk Go stacks all the time instead of just some of the time.
This includes tools like GDB, perf, and VTune. We'll improve the
situation shortly.

Change-Id: Ia40baad8f8c16aeefac05425e00b0cf478137097
Reviewed-on: https://go-review.googlesource.com/20667
Reviewed-by: Rick Hudson <rlh@golang.org>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

runtime: avoid span root marking entirely during mark termination

Currently we enqueue span root mark jobs during both concurrent mark
and mark termination, but we make the job a no-op during mark
termination.

This is silly. Instead of queueing them up just to not do them, don't
queue them up in the first place.

Change-Id: Ie1d36de884abfb17dd0db6f0449a2b7c997affab
Reviewed-on: https://go-review.googlesource.com/20666
Reviewed-by: Rick Hudson <rlh@golang.org>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

runtime: free dead G stacks concurrently

Currently we free cached stacks of dead Gs during STW stack root
marking. We do this during STW because there's no way to take
ownership of a particular dead G, so attempting to free a dead G's
stack during concurrent stack root marking could race with reusing
that G.

However, we can do this concurrently if we take a completely different
approach. One way to prevent reuse of a dead G is to remove it from
the free G list. Hence, this adds a new fixed root marking task that
simply removes all Gs from the list of dead Gs with cached stacks,
frees their stacks, and then adds them to the list of dead Gs without
cached stacks.

This is also a necessary step toward rescanning only dirty stacks,
since it eliminates another task from STW stack marking.

Change-Id: Iefbad03078b284a2e7bf30fba397da4ca87fe095
Reviewed-on: https://go-review.googlesource.com/20665
Reviewed-by: Rick Hudson <rlh@golang.org>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

runtime: split gfree list into with-stacks and without-stacks

Currently all free Gs are added to one list. Split this into two
lists: one for free Gs with cached stacks and one for Gs without
cached stacks.

This lets us preferentially allocate Gs that already have a stack, but
more importantly, it sets us up to free cached G stacks concurrently.

Change-Id: Idbe486f708997e1c9d166662995283f02d1eeb3c
Reviewed-on: https://go-review.googlesource.com/20664
Reviewed-by: Rick Hudson <rlh@golang.org>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

cmd/compile: a rule's line number is at its ->

Let's define the line number of a multiline rule as the line
number on which the -> appears.  This helps make the rule
cover analysis look a bit nicer.

Change-Id: I4ac4c09f2240285976590ecfd416bc4c05e78946
Reviewed-on: https://go-review.googlesource.com/22473
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>

cmd/compile: lazily initialize litbuf

Instead of eagerly creating strings like "literal 2.01" for every
lexed number in case we need to mention it in an error message, defer
this work to (*parser).syntax_error.

name      old allocs/op  new allocs/op  delta
Template      482k ± 0%      482k ± 0%  -0.12%   (p=0.000 n=9+10)
GoTypes      1.35M ± 0%     1.35M ± 0%  -0.04%  (p=0.015 n=10+10)
Compiler     5.45M ± 0%     5.44M ± 0%  -0.12%    (p=0.000 n=9+8)

Change-Id: I333b3c80e583864914412fb38f8c0b7f1d8c8821
Reviewed-on: https://go-review.googlesource.com/22480
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

cmd/dist: sort entries in zcgo.go generated file for deterministic build

This simplifies comparison of object files across different builds
by ensuring that the strings in the zcgo.go always appear in the
same order.

Change-Id: I3639ea4fd10e0d645b838d1bbb03cd33deca340e
Reviewed-on: https://go-review.googlesource.com/22478
Reviewed-by: Matthew Dempsky <mdempsky@google.com>

unicode: improve SimpleFold performance for ascii

This change significantly speeds up case-insensitive regexp matching.

benchmark                      old ns/op      new ns/op      delta
BenchmarkMatchEasy0i_32-8      2690           1473           -45.24%
BenchmarkMatchEasy0i_1K-8      80404          42269          -47.43%
BenchmarkMatchEasy0i_32K-8     3272187        2076118        -36.55%
BenchmarkMatchEasy0i_1M-8      104805990      66503805       -36.55%
BenchmarkMatchEasy0i_32M-8     3360192200     2126121600     -36.73%

benchmark                      old MB/s     new MB/s     speedup
BenchmarkMatchEasy0i_32-8      11.90        21.72        1.83x
BenchmarkMatchEasy0i_1K-8      12.74        24.23        1.90x
BenchmarkMatchEasy0i_32K-8     10.01        15.78        1.58x
BenchmarkMatchEasy0i_1M-8      10.00        15.77        1.58x
BenchmarkMatchEasy0i_32M-8     9.99         15.78        1.58x

Issue #13288

Change-Id: I94af7bb29e75d60b4f6ee760124867ab271b9642
Reviewed-on: https://go-review.googlesource.com/16943
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

gc: use AbsFileLine for deterministic binary export data

This version of the file name honors the -trimprefix flag,
which strips off variable parts like $WORK or $PWD.
The TestCgoConsistentResults test now passes.

Change-Id: If93980b054f9b13582dd314f9d082c26eaac4f41
Reviewed-on: https://go-review.googlesource.com/22444
Reviewed-by: Robert Griesemer <gri@golang.org>

cmd/compile: don't discard inlineable but empty functions with binary export format

Change-Id: I0f016fa000f949d27847d645b4cdebe68a8abf20
Reviewed-on: https://go-review.googlesource.com/22474
Run-TryBot: Robert Griesemer <gri@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

cmd/link: pass -no-pie (if supported) when creating a race-enabled executable.

Fixes #15443

Change-Id: Ia3593104fc1a4255926ae5675c25390604b44b7b
Reviewed-on: https://go-review.googlesource.com/22453
Run-TryBot: Michael Hudson-Doyle <michael.hudson@canonical.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>

cmd/link: fix gdb backtrace on architectures using a link register

Also adds TestGdbBacktrace to the runtime package.

Dwarf modifications written by Bryan Chan (@bryanpkc) who is also
at IBM and covered by the same CLA.

Fixes #14628

Change-Id: I106a1f704c3745a31f29cdadb0032e3905829850
Reviewed-on: https://go-review.googlesource.com/20193
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

cmd/compile/internal/gc: rewrite comment to avoid automated meaning

The comment says 'DΟ NΟT SUBMIT', and that text being in a file can cause
automated errors or warnings when trying to check the Go sources into other
source control systems.

(We reject that string in CL commit messages, which I've avoided here
by changing the O's to Ο's above.)

Change-Id: I6cdd57a8612ded5208f05a8bd6b137f44424a030
Reviewed-on: https://go-review.googlesource.com/22434
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>

cmd/compile: more sanity checks on rewrite rules

Make sure ops have the right number of args, set
aux and auxint only if allowed, etc.

Normalize error reporting format.

Change-Id: Ie545fcc5990c8c7d62d40d9a0a55885f941eb645
Reviewed-on: https://go-review.googlesource.com/22320
Reviewed-by: David Chase <drchase@google.com>

crypto/sha512: add s390x assembly implementation

Renames block to blockGeneric so that it can be called when the
assembly feature check fails. This means making block a var on
platforms without an assembly implementation (similar to the sha1
package).

Also adds a test to check that the fallback path works correctly
when the feature check fails.

name        old speed      new speed       delta
Hash8Bytes  7.13MB/s ± 2%  19.89MB/s ± 1%  +178.82%   (p=0.000 n=9+10)
Hash1K       121MB/s ± 1%    661MB/s ± 1%  +444.54%   (p=0.000 n=10+9)
Hash8K       137MB/s ± 0%    918MB/s ± 1%  +569.29%  (p=0.000 n=10+10)

Change-Id: Id65dd6e943f14eeffe39a904dc88065fc6a60179
Reviewed-on: https://go-review.googlesource.com/22402
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Michael Munday <munday@ca.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

net: ignore lame referral responses like libresolv

Fixes #15434.

Change-Id: Ia88b740df5418a6d3af1c29a03756f4234f388b0
Reviewed-on: https://go-review.googlesource.com/22428
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

cmd/link: correctly decode name length

The linker was incorrectly decoding type name lengths, causing
typelinks to be sorted out of order and in cases where the name was
the exact right length, linker panics.

Added a test to the reflect package that causes TestTypelinksSorted
to fail before this CL. It's not the exact failure seen in #15448
but it has the same cause: decodetype_name calculating the wrong
length.

The equivalent decoders in reflect/type.go and runtime/type.go
have the parenthesis in the right place.

Fixes #15448

Change-Id: I33257633d812b7d2091393cb9d6cc8a73e0138c8
Reviewed-on: https://go-review.googlesource.com/22403
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

cmd/compile: fix another bug in dominator computation

Here, "fix" means "replace".  The new dominator computation
is the "simple" algorithm from Lengauer and Tarjan's TOPLAS
paper, with minimal changes.

Also included is a test that tweaks the fixed error.

Change-Id: I0abdf53d5d64df1e67e4e62f55e88957045cd63b
Reviewed-on: https://go-review.googlesource.com/22401
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>

strings: use SSE4.2 in strings.Index on AMD64

Use PCMPESTRI instruction if available.

Index-4              21.1ns ± 0%  21.1ns ± 0%     ~     (all samples are equal)
IndexHard1-4          395µs ± 0%   105µs ± 0%  -73.53%        (p=0.000 n=19+20)
IndexHard2-4          300µs ± 0%   147µs ± 0%  -51.11%        (p=0.000 n=19+20)
IndexHard3-4          665µs ± 0%   665µs ± 0%     ~           (p=0.942 n=16+19)

Change-Id: I4f66794164740a2b939eb1c78934e2390b489064
Reviewed-on: https://go-review.googlesource.com/22337
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>

cmd/compile: sort import strings for canonical obj files

This is not necessary for reproduceability but it removes
differences due to imported package order between compiles
using textual vs binary export format. The packages list
tends to be very short, so it's ok doing it always for now.

Guarded with a documented (const) flag so it's trivial to
disable and remove eventually.

Also, use the same flag now to enforce parameter numbering.

Change-Id: Ie05d2490df770239696ecbecc07532ed62ccd5c0
Reviewed-on: https://go-review.googlesource.com/22445
Run-TryBot: Robert Griesemer <gri@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>

runtime: arm5, fix large-offset floating-point stores

The code sequence for large-offset floating-point stores
includes adding the base pointer to r11.  Make sure we
can interpret that instruction correctly.

Fixes build.

Fixes #15440

Change-Id: I7fe5a4a57e08682967052bf77c54e0ec47fcb53e
Reviewed-on: https://go-review.googlesource.com/22440
Reviewed-by: Michael Hudson-Doyle <michael.hudson@canonical.com>

cmd/compile: for now, keep parameter numbering in binary export format

The numbering is only required for parameters of functions/methods
with exported inlineable bodies. For now, always export parameter names
with internal numbering to minimize the diffs between assembly code
dumps of code compiled with the textual vs the binary format.

To be disabled again once the new export format is default.

Change-Id: I6d14c564e734cc5596c7e995d8851e06d5a35013
Reviewed-on: https://go-review.googlesource.com/22441
Reviewed-by: Matthew Dempsky <mdempsky@google.com>

cmd/compile: treat empty and absent struct field tags as identical

Fixes #15439.

Change-Id: I5a32384c46e20f8db6968e5a9e854c45ab262fe4
Reviewed-on: https://go-review.googlesource.com/22429
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

runtime: zero tmpbuf between len and cap

Zero the entire buffer so we don't need to
lower its capacity upon return.  This lets callers
do some appending without allocation.

Zeroing is cheap, the byte buffer requires only
4 extra instructions.

Fixes #14235

Change-Id: I970d7badcef047dafac75ac17130030181f18fe2
Reviewed-on: https://go-review.googlesource.com/22424
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>

cmd/compile: introduce bool operations.

Introduce OrB, EqB, NeqB, AndB to handle bool operations.

Change-Id: I53e4d5125a8090d5eeb4576db619103f19fff58d
Reviewed-on: https://go-review.googlesource.com/22412
Reviewed-by: Keith Randall <khr@golang.org>

spec: be more explicit about equivalence of empty string and absent field tags

Note that the spec already makes that point with a comment in the very first
example for struct field tags. This change is simply stating this explicitly
in the actual spec prose.

- gccgo and go/types already follow this rule
- the current reflect package API doesn't distinguish between absent tags
  and empty tags (i.e., there is no discoverable difference)

Fixes #15412.

Change-Id: I92f9c283064137b4c8651630cee0343720717a02
Reviewed-on: https://go-review.googlesource.com/22391
Reviewed-by: Russ Cox <rsc@golang.org>
Reviewed-by: Rob Pike <r@golang.org>

test: add test for issue 15084

The underlying issues have been fixed.
All the individual fixes have their own tests,
but it's still useful to have a plain source test.

Fixes #15084

Change-Id: I06c485a7d0716201bd57d1f3be53668dddd7ec14
Reviewed-on: https://go-review.googlesource.com/22426
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>

cmd/compile: encapsulate OSLICE* representation

As a nice side-effect, this allows us to
unify several code paths.

The terminology (low, high, max, simple slice expr,
full slice expr) is taken from the spec and
the examples in the spec.

This is a trial run. The plan, probably for Go 1.8,
is to change slice expressions to use Node.List
instead of OKEY, and to do some similar
tree structure changes for other ops.

Passes toolstash -cmp. No performance change.
all.bash passes with GO_GCFLAGS=-newexport.

Updates #15350

Change-Id: Ic1efdc36e79cdb95ae1636e9817a3ac8f83ab1ac
Reviewed-on: https://go-review.googlesource.com/22425
Reviewed-by: Matthew Dempsky <mdempsky@google.com>

cmd/go: disable failing TestGoGetInsecure

Update #15410

Change-Id: Iad3f2639aa7a67b11efc35a629e1893f7d87b957
Reviewed-on: https://go-review.googlesource.com/22430
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

debug/pe: better error messages

Updates #15345

Change-Id: Iae35d3e378cbc8157ba1ff91e4971ed4515a5e5c
Reviewed-on: https://go-review.googlesource.com/22394
Reviewed-by: David Crawshaw <crawshaw@golang.org>
Run-TryBot: Alex Brainman <alex.brainman@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

cmd/compile: minor cleanup in inl

* Make budget an int32 to avoid needless conversions.
* Introduce some temporary variables to reduce repetition.
* If ... args are present, they will be the last argument
  to the function. No need to scan all arguments.

Passes toolstash -cmp.

Change-Id: I55203609f5d2f25a4e238cd48c63214651120cfc
Reviewed-on: https://go-review.googlesource.com/22421
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

cmd/compile: replace len(Nodes.Slice()) with Nodes.Len()

Generated with eg:

func before(n gc.Nodes) int { return len(n.Slice()) }
func after(n gc.Nodes) int  { return n.Len() }

Change-Id: Ifdf01915e60069166afe96aa7b1d08720bf62fc5
Reviewed-on: https://go-review.googlesource.com/22420
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

test: add test that required algs get generated

This is a follow-up to CLs 19769 and 19770.

Change-Id: Ia9b71055613b80df4ce62b34fcc4f479f04f72fd
Reviewed-on: https://go-review.googlesource.com/22399
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Crawshaw <crawshaw@golang.org>

cmd/compile: use gc.Etype's String method

Passes toolstash -cmp.

Change-Id: I42c962cc5a3ffec2969f223cf238c2fdadbf5857
Reviewed-on: https://go-review.googlesource.com/22381
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

cmd/compile: give gc.Op a String method, use it

Passes toolstash -cmp.

Change-Id: I915e76374fd64aa2597e6fa47e4fa95ca00ca643
Reviewed-on: https://go-review.googlesource.com/22380
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Crawshaw <crawshaw@golang.org>

cmd/compile: don't generate algs for map buckets

Note that this is only safe because
the compiler generates multiple distinct
gc.Types. If we switch to having canonical
gc.Types, then this will need to be updated
to handle the case in which the user uses both
map[T]S and also map[[8]T]S. In that case,
the runtime needs algs for [8]T, but this could
mark the sole [8]T type as Noalg. This is a general
problem with having a single bool to represent
whether alg generation is needed for a type.

Cuts 5k off cmd/go and 22k off golang.org/x/tools/cmd/godoc,
approx 0.04% and 0.12% respectively.

For #6853 and #9930

Change-Id: I30a15ec72ecb62e2aa053260a7f0f75015fc0ade
Reviewed-on: https://go-review.googlesource.com/19769
Reviewed-by: David Crawshaw <crawshaw@golang.org>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

cmd/compile: don't generate algs for ... args

Note that this is only safe because
the compiler generates multiple distinct
gc.Types. If we switch to having canonical
gc.Types, then this will need to be updated
to handle the case in which the user uses both
map[[n]T]S and also calls a function f(...T) with n arguments.
In that case, the runtime needs algs for [n]T, but this could
mark the sole [n]T type as Noalg. This is a general
problem with having a single bool to represent
whether alg generation is needed for a type.

Cuts 17k off cmd/go and 13k off golang.org/x/tools/cmd/godoc,
approx 0.14% and 0.07% respectively.

For #6853 and #9930

Change-Id: Iccb6b9fd88ade5497d7090528a903816d340bf0a
Reviewed-on: https://go-review.googlesource.com/19770
Reviewed-by: David Crawshaw <crawshaw@golang.org>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

cmd/compile: reorder how slicelit initializes a slice

  func f(x, y, z *int) {
    a := []*int{x,y,z}
    ...
  }

We used to use:
  var tmp [3]*int
  a := tmp[:]
  a[0] = x
  a[1] = y
  a[2] = z

Now we do:
  var tmp [3]*int
  tmp[0] = x
  tmp[1] = y
  tmp[2] = z
  a := tmp[:]

Doesn't sound like a big deal, but the compiler has trouble
eliminating write barriers when using the former method because it
doesn't know that the slice points to the stack.  In the latter
method, the compiler knows the array is on the stack and as a result
doesn't emit any write barriers.

This turns out to be extremely common when building ... args, like
for calls fmt.Printf.

Makes go binaries ~1% smaller.

Doesn't have a measurable effect on the go1 fmt benchmarks,
unfortunately.

Fixes #14263
Update #6853

Change-Id: I9074a2788ec9e561a75f3b71c119b69f304d6ba2
Reviewed-on: https://go-review.googlesource.com/22395
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>

internal/trace: fix event ordering for coarse timestamps

Arm arch uses coarse-grained kernel timer as cputicks.
As the result sort.Sort smashes trace entirely. Use sort.Stable instead.

Change-Id: Idfa017a86a489be58cf239f7fe56d7f4b66b52a9
Reviewed-on: https://go-review.googlesource.com/22317
Run-TryBot: Dmitry Vyukov <dvyukov@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>

runtime/trace: test detection of broken timestamps

On some processors cputicks (used to generate trace timestamps)
produce non-monotonic timestamps. It is important that the parser
distinguishes logically inconsistent traces (e.g. missing, excessive
or misordered events) from broken timestamps. The former is a bug
in tracer, the latter is a machine issue.

Test that (1) parser does not return a logical error in case of
broken timestamps and (2) broken timestamps are eventually detected
and reported.

Change-Id: Ib4b1eb43ce128b268e754400ed8b5e8def04bd78
Reviewed-on: https://go-review.googlesource.com/21608
Reviewed-by: Austin Clements <austin@google.com>

debug/pe: introduce File.COFFSymbols and (*COFFSymbol).FullName

Reloc.SymbolTableIndex is an index into symbol table. But
Reloc.SymbolTableIndex cannot be used as index into File.Symbols,
because File.Symbols slice has Aux lines removed as it is built.

We cannot change the way File.Symbols works, so I propose we
introduce new File.COFFSymbols that does not have that limitation.

Also unlike File.Symbols, File.COFFSymbols will consist of
COFFSymbol. COFFSymbol matches PE COFF specification exactly,
and it is simpler to use.

Updates #15345

Change-Id: Icbc265853a472529cd6d64a76427b27e5459e373
Reviewed-on: https://go-review.googlesource.com/22336
Reviewed-by: David Crawshaw <crawshaw@golang.org>
Run-TryBot: Alex Brainman <alex.brainman@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

cmd/link: add -dumpdep flag to dump linker dependency graph

This is what led to https://golang.org/cl/20763 and
https://golang.org/cl/20765 to shrink binary sizes.

Change-Id: Id360d474e6153cfe32a525b0a720810fd113195b
Reviewed-on: https://go-review.googlesource.com/22392
Reviewed-by: David Crawshaw <crawshaw@golang.org>

cmd/compile: get rid of most byte and word insns for amd64

Now that we're using 32-bit ops for 8/16-bit logical operations
(to avoid partial register stalls), there's really no need to
keep track of the 8/16-bit ops at all.  Convert everything we
can to 32-bit ops.

This CL is the obvious stuff.  I might think a bit more about
whether we can get rid of weirder stuff like HMULWU.

The only downside to this CL is that we lose some information
about constants.  If we had source like:
  var a byte = ...
  a += 128
  a += 128
We will convert that to a += 256, when we could get rid of the
add altogether.  This seems like a fairly unusual scenario and
I'm happy with forgoing that optimization.

Change-Id: Ia7c1e5203d0d110807da69ed646535194a3efba1
Reviewed-on: https://go-review.googlesource.com/22382
Reviewed-by: Todd Neal <todd@tneal.org>

cmd/compile: combine stores into larger widths

Combine stores into larger widths when it is safe to do so.

Add clobber() function so stray dead uses do not impede the
above rewrites.

Fix bug in loads where all intermediate values depending on
a small load (not just the load itself) must have no other uses.
We really need the small load to be dead after the rewrite..

Fixes #14267

Change-Id: Ib25666cb19777f65082c76238fba51a76beb5d74
Reviewed-on: https://go-review.googlesource.com/22326
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: Todd Neal <todd@tneal.org>

runtime: use per-goroutine sequence numbers in tracer

Currently tracer uses global sequencer and it introduces
significant slowdown on parallel machines (up to 10x).
Replace the global sequencer with per-goroutine sequencer.

If we assign per-goroutine sequence numbers to only 3 types
of events (start, unblock and syscall exit), it is enough to
restore consistent partial ordering of all events. Even these
events don't need sequence numbers all the time (if goroutine
starts on the same P where it was unblocked, then start does
not need sequence number).
The burden of restoring the order is put on trace parser.
Details of the algorithm are described in the comments.

On http benchmark with GOMAXPROCS=48:
no tracing: 5026 ns/op
tracing: 27803 ns/op (+453%)
with this change: 6369 ns/op (+26%, mostly for traceback)

Also trace size is reduced by ~22%. Average event size before: 4.63
bytes/event, after: 3.62 bytes/event.

Besides running trace tests, I've also tested with manually broken
cputicks (random skew for each event, per-P skew and episodic random skew).
In all cases broken timestamps were detected and no test failures.

Change-Id: I078bde421ccc386a66f6c2051ab207bcd5613efa
Reviewed-on: https://go-review.googlesource.com/21512
Run-TryBot: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

doc: mention security from contribution guidelines

Fixes #15413

Change-Id: I837a391276eed565cf66d3715ec68b7b959ce143
Reviewed-on: https://go-review.googlesource.com/22390
Reviewed-by: Andrew Gerrand <adg@golang.org>

cmd/compile: remove redundant "// fallthrough" comments

Change-Id: Ia3f262f06592b66447c213e2350402cd5e6e2ccd
Reviewed-on: https://go-review.googlesource.com/22389
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>

cmd/compile: don't export pos info in new export format for now

Exporting filenames as part of the position information can lead
to different object files which breaks tests.

Change-Id: Ia678ab64293ebf04bf83601e6ba72919d05762a4
Reviewed-on: https://go-review.googlesource.com/22385
Reviewed-by: Matthew Dempsky <mdempsky@google.com>