Shenghou Ma [Tue, 7 Aug 2012 15:45:50 +0000 (23:45 +0800)]
runtime: inline several float64 routines to speed up complex128 division
Depends on CL 6197045.
Result obtained on Core i7 620M, Darwin/amd64:
benchmark old ns/op new ns/op delta
BenchmarkComplex128DivNormal 57 28 -50.78%
BenchmarkComplex128DivNisNaN 49 15 -68.90%
BenchmarkComplex128DivDisNaN 49 15 -67.88%
BenchmarkComplex128DivNisInf 40 12 -68.50%
BenchmarkComplex128DivDisInf 33 13 -61.06%
Result obtained on Core i7 620M, Darwin/386:
benchmark old ns/op new ns/op delta
BenchmarkComplex128DivNormal 89 50 -44.05%
BenchmarkComplex128DivNisNaN 307 802 +161.24%
BenchmarkComplex128DivDisNaN 309 788 +155.02%
BenchmarkComplex128DivNisInf 278 237 -14.75%
BenchmarkComplex128DivDisInf 46 22 -52.46%
Result obtained on 700MHz OMAP4460, Linux/ARM:
benchmark old ns/op new ns/op delta
BenchmarkComplex128DivNormal 1557 465 -70.13%
BenchmarkComplex128DivNisNaN 1443 220 -84.75%
BenchmarkComplex128DivDisNaN 1481 218 -85.28%
BenchmarkComplex128DivNisInf 952 216 -77.31%
BenchmarkComplex128DivDisInf 861 231 -73.17%
The 386 version has a performance regression, but as we have
decided to use SSE2 instead of x87 FPU for 386 too (issue 3912),
I won't address this issue.
R=dsymonds, mchaten, iant, dave, mtj, rsc, r
CC=golang-dev
https://golang.org/cl/6024045
Shenghou Ma [Tue, 7 Aug 2012 01:57:14 +0000 (09:57 +0800)]
math, runtime: use a NaN that matches gcc's
our old choice is not working properly at least on VFPv2 in
ARM1136JF-S (it's not preserved across float64->float32 conversions).
Shenghou Ma [Tue, 7 Aug 2012 01:38:35 +0000 (09:38 +0800)]
misc/cgo/{life,stdio}, test/run.go: use test/run.go to do the cgo tests
Enhances test/run.go to support testing other directories
Will enable stdio tests on Windows in a follow-up CL.
Andrew Balholm [Mon, 6 Aug 2012 23:35:09 +0000 (09:35 +1000)]
exp/html: fix foster-parenting when elements are implicitly closed
When an element (like <nobr> or <p>) was implicitly closed by another
start tag, it would keep foster parenting from working because
the check for what was on top of the stack of open elements was
in the wrong place.
Alexey Borzenkov [Sun, 5 Aug 2012 21:24:32 +0000 (17:24 -0400)]
syscall: return EINVAL when string arguments have NUL characters
Since NUL usually terminates strings in underlying syscalls, allowing
it when converting string arguments is a security risk, especially
when dealing with filenames. For example, a program might reason that
filename like "/root/..\x00/" is a subdirectory or "/root/" and allow
access to it, while underlying syscall will treat "\x00" as an end of
that string and the actual filename will be "/root/..", which might
be unexpected. Returning EINVAL when string arguments have NUL in
them makes sure this attack vector is unusable.
Russ Cox [Sun, 5 Aug 2012 18:35:35 +0000 (14:35 -0400)]
misc/dashboard/codereview: show first line of last message in thread
This line helps me to tell whether the CL is waiting for me or I'm waiting for the author.
Also:
- vertical-align table cells so buttons are always aligned with CL headers.
- add email= to show front page for someone else.
Demo at http://rsc.gocodereview.appspot.com/.
Until this is deployed for real, some recently changed CLs may be
missing the 'first line of last message' part.
Andrew Balholm [Sun, 5 Aug 2012 12:32:35 +0000 (22:32 +1000)]
exp/html: don't treat SVG <title> like HTML <title>
The content of an HTML <title> element is RCDATA, but the content of an SVG
<title> element is parsed as tags. Now the parser doesn't go into RCDATA
mode in foreign content.
Dmitriy Vyukov [Sat, 4 Aug 2012 15:02:12 +0000 (18:02 +0300)]
cmd/go: allow to use syso files with cgo
I have C functions implemented in .syso file (rather than .so or inlined in .go file).
W/o this change the gcc invocation fails with undefined symbols.
Rémy Oudompheng [Fri, 3 Aug 2012 19:47:26 +0000 (21:47 +0200)]
cmd/gc: accept switches on comparable arrays.
The compiler is incorrectly rejecting switches on arrays of
comparable types. It also doesn't catch incomparable structs
when typechecking the switch, leading to unreadable errors
during typechecking of the generated code.
Adam Langley [Fri, 3 Aug 2012 17:28:05 +0000 (13:28 -0400)]
crypto/tls: remove flakey tests
This removes some (non-short only) tests that involved connections to
the outside world and thus were flakey. This does remove some test
coverage of the root fetchers on non-Windows platforms, but the right
place for that is crypto/x509.
Marcel van Lohuizen [Fri, 3 Aug 2012 07:01:21 +0000 (09:01 +0200)]
exp/locale/collate: changed API of Builder to be more convenient
for dealing with CLDR files:
- Add now taxes a list of indexes of colelems that are variables. Checking and
handling is now done by the Builder. VariableTop is now also properly generated
using the Build method.
- Introduced separate Builder, called Tailoring, for creating tailorings of root
table. This clearly separates the functionality for building a table based on
weights (the allkeys* files) versus tables based on LDML XML files.
- Tailorings are now added by two calls instead of one: SetAnchor and Insert.
This more closely reflects the structure of LDML side and simplifies the
implementation of both the client and library side. It also preserves
some information that is otherwise hard to recover for the Builder.
- Allow the LDML XML element extend to be passed to Insert. This simplifies
both client and library implementation.
Shenghou Ma [Wed, 1 Aug 2012 03:26:33 +0000 (23:26 -0400)]
test/run.go: fix compiledir test on windows
we can't import "./bug0" on windows, as it will trigger
"import path contains invalid character ':'" error.
instead, we pass "-D." and "-I." to gc to override this
behavior. this idea is due to remyoudompheng.
Andrew Balholm [Tue, 31 Jul 2012 23:35:02 +0000 (09:35 +1000)]
exp/html: tokenize attributes of end tags
If an end tag has an attribute that is a quoted string containing '>',
the tokenizer would end the tag prematurely. Now it reads the attributes
on end tags just as it does on start tags, but the high-level interface
still doesn't return them, because their presence is a parse error.
Shenghou Ma [Mon, 30 Jul 2012 22:44:48 +0000 (18:44 -0400)]
cmd/cgo: use 1 as last entry for __cgodebug_data
LLVM-based gcc will place all-zero data in a zero-filled
section, but our debug/macho can't handle that.
Fixes #3821.
Rob Pike [Mon, 30 Jul 2012 22:11:20 +0000 (15:11 -0700)]
text/template/parse: fix data race
The situation only affects diagnostics but is easy to fix.
When computing lineNumber, use the position of the last item
returned by nextItem rather than the current state of the lexer.
This is internal only and does not affect the API.
The compiledir pattern compiles all files xxx.dir/*.go
in lexicographic order (which is assumed to coincide with
the topological order of dependencies).
misc/dashboard/codereview: add LastUpdate field to CL
To be filled in by a later CL. I deployed a test version to
App Engine to work on setting the values, so there are
a few records that have this field set already.
That field is breaking the live version, so I have pushed
a new copy with this 1-line change to the live version
I assumed that appengine/datastore was like every other
marshaling and unmarshaling package we have in Go
(for example, encoding/gob, encoding/json, encoding/xml,
and protobuf) and that if it loaded an unknown field it would
just ignore it. Apparently not. Sorry.
This fixes a data race (usually just harmlessly updating
"GET" to "GET"), but also follows RFC 2616 Sec 5.1.1 which
says that the request method is case-sensitive.
Darwin kernels have a bug in accept() where error result from
an internal call is not checked and socket is accepted instead
of ECONNABORTED error. However, such sockets have no sockaddr,
which results in EAFNOSUPPORT error from anyToSockaddr, making
Go http servers running on Mac OS X easily susceptible to
denial of service from simple port scans with nmap.
Fixes #3849.
David Symonds [Sun, 29 Jul 2012 22:48:51 +0000 (08:48 +1000)]
misc/vim: fix :Import insertion heuristic.
If a factored import group has a blank line, assume it is dividing
separate groups of imports (e.g. standard library vs. site-specific).
import (
"bytes"
"io"
"mycorp/package"
)
The most common case is inserting new standard library imports,
which are usually (stylistically) the first group, so we should drop
"net" in the above example immediately after "io".
Since this logic is getting non-trivial, add a test.
exp/locale/collate: changed trie in first step towards support for multiple locales.
- Allow handles into the trie for different locales. Multiple tables share the same
try to allow for reuse of blocks.
- Significantly improved memory footprint and reduced allocations of trieNodes.
This speeds up generation by about 30% and allows keeping trieNodes around
for multiple locales during generation.
- Renamed print method to fprint.
Ian Lance Taylor [Fri, 27 Jul 2012 05:46:20 +0000 (22:46 -0700)]
runtime: ignore signal 33 == SIGSETXID on GNU/Linux
When a cgo program calls setuid, setgid, etc., the GNU/Linux
pthread library sends signal SIGSETXID to each thread to tell
it to update its UID info. If Go is permitted to intercept
the default SIGSETXID signal handler, the program will hang.
This patch tells the runtime package to not try to intercept
SIGSETXID on GNU/Linux. This will be odd if a Go program
wants to try to use that signal, but it means that cgo
programs that call setuid, etc., won't hang.
Robert Griesemer [Fri, 27 Jul 2012 00:09:11 +0000 (17:09 -0700)]
go/printer: don't crash if ast.FuncType.Params is nil
The go/ast comment for FuncType.Params says that the field may be nil.
Make sure the printer accepts such a value. The go/parser always sets
the field (to provide parenthesis position information), but a program
creating a Go AST from scatch may not.
Added corresponding test case.
Fixes #3870.
R=golang-dev, r
CC=golang-dev
https://golang.org/cl/6448060