Alan Donovan [Tue, 29 Jan 2013 15:49:16 +0000 (10:49 -0500)]
exp/ssa: make Parameters values, not addresses.
We explicitly spill all parameters to the frame during initial
SSA construction. (Later passes will remove spills.)
We now properly handle local Allocs escaping via Captures.
Dmitriy Vyukov [Tue, 29 Jan 2013 10:57:11 +0000 (14:57 +0400)]
runtime: dump the full stack of a throwing goroutine
Useful for debugging of runtime bugs.
+ Do not print "stack segment boundary" unless GOTRACEBACK>1.
+ Do not traceback system goroutines unless GOTRACEBACK>1.
Carl Shapiro [Mon, 28 Jan 2013 23:47:25 +0000 (15:47 -0800)]
cmd/ld: avoid a segfault when dumping the symbol table
The dumping routine incorrectly assumed that all incoming
symbols would be non-nil and load through it to retrieve the
symbol name. Instead of using the symbol to retrieve a name,
use the name provided by the caller.
Dave Cheney [Mon, 28 Jan 2013 10:05:25 +0000 (21:05 +1100)]
src: add race.bash
Add race.bash so anyone with suitable hardware can run a race detector build. race.bash can be called from the dashboard builder by passing -cmd="race.bash".
Original source for race.bash is here, http://code.google.com/p/go-wiki/wiki/DashboardBuilders
Ian Lance Taylor [Sun, 27 Jan 2013 02:16:43 +0000 (18:16 -0800)]
runtime: use new CNT_MASK in lfstack
This is for SPARC64, a 64-bit processor that uses all 64-bits
of virtual addresses. The idea is to use the low order 3 bits
to at least get a small ABA counter. That should work since
pointers are aligned. The idea is for SPARC64 to set CNT_MASK
== 7, PTR_BITS == 0, PTR_MASK == 0xffffffffffffff8.
Also add uintptr casts to avoid GCC warnings. The gccgo
runtime code is compiled with GCC, and GCC warns when casting
between a pointer and a type of a different size.
John Graham-Cumming [Fri, 25 Jan 2013 18:20:19 +0000 (10:20 -0800)]
net/http: fix Content-Length/Transfer-Encoding on HEAD requests
net/http currently assumes that the response to a HEAD request
will always have a Content-Length header. This is incorrect.
RFC2616 says: "The HEAD method is identical to GET except that
the server MUST NOT return a message-body in the response. The
metainformation contained in the HTTP headers in response to a
HEAD request SHOULD be identical to the information sent in
response to a GET request. This method can be used for
obtaining metainformation about the entity implied by the
request without transferring the entity-body itself. This
method is often used for testing hypertext links for validity,
accessibility, and recent modification."
This means that three cases are possible: a Content-Length
header, a Transfer-Encoding header or neither. In the wild the
following sites exhibit these behaviours (curl -I):
HEAD on http://www.google.co.uk/ has Transfer-Encoding: chunked
HEAD on http://www.bbc.co.uk/ has Content-Length: 45247
HEAD on http://edition.cnn.com/ has neither header
This patch does not remove the ErrMissingContentLength error
for compatibility reasons, but it is no longer used.
Andrew Gerrand [Thu, 24 Jan 2013 23:06:18 +0000 (10:06 +1100)]
misc/dashboard/builder: synchronize accesses to goroot, always -commit
This prevents the occasional issue when Mercurial screws up the locking
itself, and by moving the locking into this process we can use the
goroot for other things (such as automatically updating the builder
binary).
It also asks all builders to poll for new commits.
Alan Donovan [Thu, 24 Jan 2013 19:22:17 +0000 (14:22 -0500)]
go/types: expose types.IsIdentical, the Type equivalence relation.
This function is absolutely critical for clients such as
exp/ssa, and too complex for clients to duplicate.
As with CL 7200046, gri expressed in the doc below [gophers
only] before going on leave that he intended to expose such a
predicate, though his wording suggests as an interface method
of Type rather than a standalone function. (My preference is
for binary methods to be standalone; see "On Binary Methods",
Kim Bruce, 1995). In any case if he wishes to move it that's
easily accommodated by clients.
Alan Donovan [Thu, 24 Jan 2013 19:21:51 +0000 (14:21 -0500)]
go/types: add String() method to Type interface.
All implementations delegate to typeString.
Though I don't wish to exploit gri's absence to change
his code, this change is pretty low-risk and he assented to it
in the blue ink in the doc below [gophers only].
https://docs.google.com/a/google.com/document/d/1-DQ4fxlMDs9cYtnkKhAAehX6MArjOQyJsRXp-6kiJLA/edit#
Mikkel Krautz [Wed, 23 Jan 2013 17:20:17 +0000 (01:20 +0800)]
crypto/x509: skip SystemRootsError test on Windows
On Windows, crypto/x509 passes through to Windows's CryptoAPI
to verify certificate chains. This method can't produce a
SystemRootsError, so make sure we always skip the test on
Windows.
This is needed because testVerify is called in both
TestGoVerify and TestSystemVerify on Windows - one is for
testing the Go verifier, the other one is for testing the
CryptoAPI verifier. The orignal CL tried to sidestep
this issue by setting systemSkip to true, but that only
affected TestSystemVerify.
Marcel van Lohuizen [Wed, 23 Jan 2013 13:15:51 +0000 (14:15 +0100)]
exp/locale/collate: preparation for adding Search API. Also changed the collate API
further to how (I believe) it will end up being.
It is nicer to separate search from sorting functionality. Collation needs tables that
are not needed by search and vice-versa. The common functionality is separated out
in the Weigher interface. As this interface is very low-level, it will be moved to
a sub package (colltab) in a next CL.
The types that will move to this package are Weigher, Elem, and Level. The addition
of Elem allows for removing some of the duplicate code between collate and collate/build.
This CL also introduces some stubs for a higher-level API for options. The default
proposed options are quite complex and require the user to have a decent understanding
of Unicode collation. The new options hide a lot of the complexity.
Akshat Kumar [Wed, 23 Jan 2013 03:42:44 +0000 (22:42 -0500)]
syscall: fix fork-exec/wait inconsistencies for Plan 9
Fixes the fork-exec/wait race condition for ForkExec
as well, by making it use startProcess. This makes the
comment for StartProcess consistent as well.
Further, the passing of Waitmsg data in startProcess
and WaitProcess is protected against possible forks
from outside of ForkExec and StartProcess, which might
cause interference with the Await call.
Andrew Gerrand [Wed, 23 Jan 2013 01:17:11 +0000 (12:17 +1100)]
net/url: use bytes.Buffer in (*URL).String
BenchmarkString before:
11990 ns/op 1621 B/op 73 allocs/op
Using bytes.Buffer:
8774 ns/op 1994 B/op 40 allocs/op
I also tried making a version of escape() that writes directly to the
bytes.Buffer, but it only saved 1 alloc/op and increased CPU time by
about 10%. Didn't seem worth the extra code path.
Andrew Gerrand [Wed, 23 Jan 2013 00:37:06 +0000 (11:37 +1100)]
net/url: generate correct Path when hostname empty
Parse("file:///foo") previously returned a URL with Scheme "file"
and Path "///foo". Now it returns a URL with Path "/foo",
such that
&URL{Scheme: "file", Path: "/foo"}.String() == "file:///foo"
This means that parsing and stringifying the URL "file:/foo"
returns "file:///foo", technically a regression but one that only
affects a corner case.
Dave Cheney [Tue, 22 Jan 2013 23:22:33 +0000 (10:22 +1100)]
testing: add Skip/Skipf
This proposal adds two methods to *testing.T, Skip(string) and Skipf(format, args...). The intent is to replace the existing log and return idiom which currently has 97 cases in the standard library. A simple example of Skip would be:
func TestSomethingLong(t *testing.T) {
if testing.Short() {
t.Skip("skipping test in short mode.")
// not reached
}
... time consuming work
}
Additionally tests can be skipped anywhere a *testing.T is present. An example adapted from the go.crypto/ssh/test package would be:
// setup performs some before test action and returns a func()
// which should be defered by the caller for cleanup.
func setup(t *testing.T) func() {
...
cmd := exec.Command("sshd", "-f", configfile, "-i")
if err := cmd.Run(); err != nil {
t.Skipf("could not execute mock ssh server: %v", err)
}
...
return func() {
// stop subprocess and cleanup
}
}
Rick Arnold [Tue, 22 Jan 2013 22:49:07 +0000 (17:49 -0500)]
encoding/json: ignore unexported fields in Unmarshal
Go 1.0 behavior was to create an UnmarshalFieldError when a json value name matched an unexported field name. This error will no longer be created and the field will be skipped instead.
Anthony Martin [Tue, 22 Jan 2013 22:05:13 +0000 (17:05 -0500)]
cmd/go: suppress extraneous newlines in list
Before:
$ go list -f '{{range .Deps}}{{println $.Name .}}{{end}}' math time
math runtime
math unsafe
time errors
time runtime
time sync
time sync/atomic
time syscall
time unsafe
$
After:
$ go list -f '{{range .Deps}}{{println $.Name .}}{{end}}' math time
math runtime
math unsafe
time errors
time runtime
time sync
time sync/atomic
time syscall
time unsafe
$
Robin Eklind [Tue, 22 Jan 2013 21:44:35 +0000 (13:44 -0800)]
mime, strconv: Make testdata more consistent.
All packages place testdata in a specific directory with the name
"testdata". The mime and strconv packages have been updated to use
the same convention.
mime: Move "mime/test.types" to "mime/testdata/test.types". Update test
code accordingly.
strconv: Move "strconv/testfp.txt" to "strconv/testdata/testfp.txt".
Update test code accordingly.
Adam Langley [Tue, 22 Jan 2013 15:10:38 +0000 (10:10 -0500)]
crypto/tls: allow the server to enforce its ciphersuite preferences.
Previously, Go TLS servers always took the client's preferences into
account when selecting a ciphersuite. This change adds the option of
using the server's preferences, which can be expressed by setting
tls.Config.CipherSuites.
This mirrors Apache's SSLHonorCipherOrder directive.
Dmitriy Vyukov [Tue, 22 Jan 2013 09:44:49 +0000 (13:44 +0400)]
runtime: account stop-the-world time in the "other" GOGCTRACE section
Currently it's summed to mark phase.
The change makes it easier to diagnose long stop-the-world phases.
R=golang-dev, bradfitz, dave
CC=golang-dev
https://golang.org/cl/7182043
Shenghou Ma [Mon, 21 Jan 2013 18:50:27 +0000 (02:50 +0800)]
cmd/5l: move offset2 into Adr.u0 union to save 4/8 bytes for Adr/Prog resp.
sizeof(Adr) from 24 bytes down to 20 bytes.
sizeof(Prog) from 84 bytes down to 76 bytes.
5l linking cmd/godoc statistics:
Before:
Maximum resident set size (kbytes): 106668
After:
Maximum resident set size (kbytes): 99412
R=golang-dev, dave
CC=golang-dev
https://golang.org/cl/7100059
Shenghou Ma [Mon, 21 Jan 2013 18:48:40 +0000 (02:48 +0800)]
text/template/parse: don't show itemType in error messages
so that the user don't need to decipher something like this:
template: main:1: expected %!s(parse.itemType=14) in end; got "|"
now they get this:
template: main:1: unexpected "|" in end
R=golang-dev, r
CC=golang-dev
https://golang.org/cl/7128054
Adam Langley [Mon, 21 Jan 2013 16:22:08 +0000 (11:22 -0500)]
crypto/cipher: don't persist errors in StreamWriter.
I messed this up from the beginning. The receiver isn't a pointer so
setting Err is useless. In order to maintain the API, just remove the
superfluous code.
Anthony Martin [Sat, 19 Jan 2013 03:08:00 +0000 (19:08 -0800)]
cmd/8g, cmd/dist, cmd/gc: fix warnings on Plan 9
cmd/8g/gsubr.c: unreachable code
cmd/8g/reg.c: overspecifed class
cmd/dist/plan9.c: unused parameter
cmd/gc/fmt.c: stkdelta is now a vlong
cmd/gc/racewalk.c: used but not set
Anthony Martin [Sat, 19 Jan 2013 03:00:38 +0000 (19:00 -0800)]
cmd/6c, cmd/8c: fix print format for Prog
The FmtLong flag should only be used with the %D verb
when printing an ATEXT Prog. It was erroneously used
for every Prog except ADATA. This caused a preponderance
of exclamation points, "!!", in the assembly listings.
I also cleaned up the code so that the list.c files look
very similar. Now the real differences are easily spotted
with a simple diff.
Matthew Dempsky [Fri, 18 Jan 2013 22:29:53 +0000 (17:29 -0500)]
cmd/6c: Optimize rotate expressions to use rotate instructions.
For simplicity, only recognizes expressions of the exact form
"(x << a) | (x >> b)" where x is a variable and a and b are
integer constant expressions that add to x's bit width.
Fixes #4629.
$ cat rotate.c
unsigned int
rotate(unsigned int x)
{
x = (x << 3) | (x >> (sizeof(x) * 8 - 3));
return x;
}
## BEFORE
$ go tool 6c -S rotate.c
(rotate.c:2) TEXT rotate+0(SB),$0-8
(rotate.c:2) MOVL x+0(FP),!!DX
(rotate.c:4) MOVL DX,!!AX
(rotate.c:4) SALL $3,!!AX
(rotate.c:4) MOVL DX,!!CX
(rotate.c:4) SHRL $29,!!CX
(rotate.c:4) ORL CX,!!AX
(rotate.c:5) RET ,!!
(rotate.c:5) RET ,!!
(rotate.c:5) END ,!!
## AFTER
$ go tool 6c -S rotate.c
(rotate.c:2) TEXT rotate+0(SB),$0-8
(rotate.c:4) MOVL x+0(FP),!!AX
(rotate.c:4) ROLL $3,!!AX
(rotate.c:5) RET ,!!
(rotate.c:5) RET ,!!
(rotate.c:5) END ,!!
Akshat Kumar [Fri, 18 Jan 2013 21:43:25 +0000 (16:43 -0500)]
syscall, os: fix a fork-exec/wait race in Plan 9.
On Plan 9, only the parent of a given process can enter its wait
queue. When a Go program tries to fork-exec a child process
and subsequently waits for it to finish, the goroutines doing
these two tasks do not necessarily tie themselves to the same
(or any single) OS thread. In the case that the fork and the wait
system calls happen on different OS threads (say, due to a
goroutine being rescheduled somewhere along the way), the
wait() will either return an error or end up waiting for a
completely different child than was intended.
This change forces the fork and wait syscalls to happen in the
same goroutine and ties that goroutine to its OS thread until
the child exits. The PID of the child is recorded upon fork and
exit, and de-queued once the child's wait message has been read.
The Wait API, then, is translated into a synthetic implementation
that simply waits for the requested PID to show up in the queue
and then reads the associated stats.
Rémy Oudompheng [Fri, 18 Jan 2013 21:40:32 +0000 (22:40 +0100)]
cmd/gc: fix handling of struct padding in hash/eq.
The test case of issue 4585 was not passing due to
miscalculation of memequal args, and the previous fix
does not handle padding at the end of a struct.
Handling of padding at end of structs also fixes the case
of [n]T where T is such a padded struct.
Russ Cox [Fri, 18 Jan 2013 20:10:36 +0000 (15:10 -0500)]
build: change GO386=sse to GO386=sse2
sse2 is a more precise description of the requirement,
and it matches what people will see in, for example
grep sse2 /proc/cpuinfo # linux
sysctl hw.optional.sse2 # os x