]> Cypherpunks repositories - gostls13.git/log
gostls13.git
9 years ago[dev.ssa] cmd/compile: be safer about uintptr/unsafe.Pointer conversions
Keith Randall [Tue, 10 Nov 2015 23:35:36 +0000 (15:35 -0800)]
[dev.ssa] cmd/compile: be safer about uintptr/unsafe.Pointer conversions

Make sure that when a pointer value is live across a function
call, we save it as a pointer.  (And similarly a uintptr
live across a function call should not be saved as a pointer.)

Add a nasty test case.

This is probably what is preventing the merge from master
to dev.ssa.  Signs point to something like this bug happening
in mallocgc.

Change-Id: Ib23fa1251b8d1c50d82c6a448cb4a4fc28219029
Reviewed-on: https://go-review.googlesource.com/16830
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
9 years ago[dev.ssa] cmd/compile: Deduplicate panic{index,slice,divide} calls
Keith Randall [Tue, 10 Nov 2015 05:35:40 +0000 (21:35 -0800)]
[dev.ssa] cmd/compile: Deduplicate panic{index,slice,divide} calls

Panics are only distinguished by their type and line number, so
if we can trigger two of those panics in the same line, use the
same panic call.  For example, in a[i]+b[j] we need only one
panicindex call that both bounds checks can use.

Change-Id: Ia2b6d3b1a67f2775df05fb72b8a1b149833572b7
Reviewed-on: https://go-review.googlesource.com/16772
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: David Chase <drchase@google.com>
9 years ago[dev.ssa] cmd/compile: some SSA optimizations
Keith Randall [Tue, 10 Nov 2015 04:54:34 +0000 (20:54 -0800)]
[dev.ssa] cmd/compile: some SSA optimizations

Some optimizations of things I've seen looking at generated code.
  (x+y)-x == y
  x-0 == x
The ptr portion of the constant string "" can be nil.

Also update TODO with recent changes.

Change-Id: I02c41ca2f9e9e178bf889058d3e083b446672dbe
Reviewed-on: https://go-review.googlesource.com/16771
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: David Chase <drchase@google.com>
9 years ago[dev.ssa] cmd/compile: Handle ONOT in conditionals as well
Keith Randall [Tue, 3 Nov 2015 17:30:17 +0000 (09:30 -0800)]
[dev.ssa] cmd/compile: Handle ONOT in conditionals as well

Might as well, for a && !(b && c) and the like.

Change-Id: I2548b6e6ee5870e074bcef6edd56a7db6e81d70f
Reviewed-on: https://go-review.googlesource.com/16600
Reviewed-by: David Chase <drchase@google.com>
Run-TryBot: David Chase <drchase@google.com>

9 years ago[dev.ssa] cmd/compile: start arguments as spilled
Keith Randall [Mon, 2 Nov 2015 16:10:26 +0000 (08:10 -0800)]
[dev.ssa] cmd/compile: start arguments as spilled

Declare a function's arguments as having already been
spilled so their use just requires a restore.

Allow spill locations to be portions of larger objects the stack.
Required to load portions of compound input arguments.

Rename the memory input to InputMem.  Use Arg for the
pre-spilled argument values.

Change-Id: I8fe2a03ffbba1022d98bfae2052b376b96d32dda
Reviewed-on: https://go-review.googlesource.com/16536
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
9 years ago[dev.ssa] cmd/compile: Do pointer arithmetic with int, not uintptr
Keith Randall [Tue, 3 Nov 2015 05:28:13 +0000 (21:28 -0800)]
[dev.ssa] cmd/compile: Do pointer arithmetic with int, not uintptr

Be more consistent about this.  There's no reason to do the
pointer arithmetic on a different type, as sizeof(int) >=
sizeof(ptr) on all of our platforms.  It simplifies our
rewrite rules also, except for a few that need duplication.

Add some more constant folding to get constant indexing and
slicing to fold down to nothing.

Change-Id: I3e56cdb14b3dc1a6a0514f0333e883f92c19e3c7
Reviewed-on: https://go-review.googlesource.com/16586
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: David Chase <drchase@google.com>
9 years ago[dev.ssa] cmd/compile: better ANDAND and OROR in IF and FOR
Keith Randall [Tue, 3 Nov 2015 00:56:53 +0000 (16:56 -0800)]
[dev.ssa] cmd/compile: better ANDAND and OROR in IF and FOR

For the statement

    if a && b { target }

the old code allocated a new variable v and did:

    v = a
    if a {
       v = b
    }
    if v { goto target }

The new code does:

    if a {
      if b { goto target }
    }

The new arrangement tends to generate much more efficient code.  In
particular, there is no temporary variable and there is only one join
point instead of two.

The old code is still used for ANDAND and OROR which are not
direct descendents of IF or FOR statements.

Change-Id: I082f246d27c823c6f32d1287300e4b0911607507
Reviewed-on: https://go-review.googlesource.com/16584
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: David Chase <drchase@google.com>
9 years ago[dev.ssa] cmd/compile: default compile+test with SSA
David Chase [Mon, 26 Oct 2015 21:34:06 +0000 (17:34 -0400)]
[dev.ssa] cmd/compile: default compile+test with SSA

Some tests disabled, some bifurcated into _ssa and not,
with appropriate logging added to compiler.

"tests/live.go" in particular needs attention.

SSA-specific testing removed, since it's all SSA now.

Added "-run_skips" option to tests/run.go to simplify
checking whether a test still fails (or how it fails)
on a skipped platform.

The compiler now compiles with SSA by default.
If you don't want SSA, specify GOSSAHASH=n (or N) as
an environment variable.  Function names ending in "_ssa"
are always SSA-compiled.

GOSSAFUNC=fname retains its "SSA for fname, log to ssa.html"
GOSSAPKG=pkg only has an effect when GOSSAHASH=n
GOSSAHASH=10101 etc retains its name-hash-matching behavior
for purposes of debugging.

See #13068

Change-Id: I8217bfeb34173533eaeb391b5f6935483c7d6b43
Reviewed-on: https://go-review.googlesource.com/16299
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: David Chase <drchase@google.com>

9 years ago[dev.ssa] cmd/compile/internal/ssa: rewrite rules for const comparisons
Todd Neal [Tue, 27 Oct 2015 22:46:53 +0000 (17:46 -0500)]
[dev.ssa] cmd/compile/internal/ssa: rewrite rules for const comparisons

Shaves ~8k off the go binary on darwin x64.

Change-Id: I73396af44ae28cd4cfc675290d6858f304d45b76
Reviewed-on: https://go-review.googlesource.com/16456
Run-TryBot: Todd Neal <todd@tneal.org>
Reviewed-by: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years ago[dev.ssa] cmd/compile: flag recomputing: find original values correctly
Keith Randall [Thu, 29 Oct 2015 20:41:02 +0000 (13:41 -0700)]
[dev.ssa] cmd/compile: flag recomputing: find original values correctly

We "spill" flag values by recomputing them from their original
inputs.  The "find original inputs" part of the algorithm was
a hack.  It was broken by rematerialization.  This change does
the real job of keeping track of original values for each
spill/restore/flagrecompute/rematerialization we issue.

Change-Id: I95088326a4ee4958c98148b063e518c80e863e4c
Reviewed-on: https://go-review.googlesource.com/16500
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years ago[dev.ssa] cmd/compile/internal/ssa: reuse Aux values for PEXTERN
Todd Neal [Wed, 28 Oct 2015 02:35:48 +0000 (21:35 -0500)]
[dev.ssa] cmd/compile/internal/ssa: reuse Aux values for PEXTERN

This improves cse and works correctly now that divide by zero is checked
explicitly.

Change-Id: If54fbe403ed5230b897afc5def644ba9f0056dfd
Reviewed-on: https://go-review.googlesource.com/16454
Run-TryBot: Todd Neal <todd@tneal.org>
Reviewed-by: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years ago[dev.ssa] cmd/compile: split op rewrites into separate functions
Keith Randall [Tue, 27 Oct 2015 04:49:31 +0000 (21:49 -0700)]
[dev.ssa] cmd/compile: split op rewrites into separate functions

The single value rewrite function is too big.  Some compilers
fail on it (out of memory, branch offset too large).  Break it
up into a rewrite function per op.

Change-Id: Iede697c8a1a3a22b485cd0dc85d3e233160c89c2
Reviewed-on: https://go-review.googlesource.com/16347
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Todd Neal <todd@tneal.org>
9 years ago[dev.ssa] cmd/compile: fix printing of live information
Keith Randall [Wed, 28 Oct 2015 17:40:47 +0000 (10:40 -0700)]
[dev.ssa] cmd/compile: fix printing of live information

SSA generates ACALL assembly with the target in a *Sym.
The old compiler generates both that *Sym and a *Node.
Use the *Sym to print the live info so it works with both compilers.

Change-Id: I0b12a161f83e76638604358c21b9f5abb31ce950
Reviewed-on: https://go-review.googlesource.com/16432
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: David Chase <drchase@google.com>
9 years ago[dev.ssa] cmd/compile: make zero-divide panic from div/mod explicit
David Chase [Wed, 28 Oct 2015 17:55:46 +0000 (13:55 -0400)]
[dev.ssa] cmd/compile: make zero-divide panic from div/mod explicit

Added an explicit compare-zero and branch-to-panic for
integer division and mod so that other optimizations will
not be fooled by their implicit panics.

Change-Id: Ibf96f636b541c0088861907c537a6beb4b99fa4c
Reviewed-on: https://go-review.googlesource.com/16450
Reviewed-by: Keith Randall <khr@golang.org>
9 years ago[dev.ssa] cmd/compile: remember names of values
Keith Randall [Thu, 22 Oct 2015 21:22:38 +0000 (14:22 -0700)]
[dev.ssa] cmd/compile: remember names of values

For debugging, spill values to named variables instead of autotmp_
variables if possible.  We do this by keeping a name -> value map
for each function, keep it up-to-date during deadcode elim, and use
it to override spill decisions in stackalloc.

It might even make stack frames a bit smaller, as it makes it easy
to identify a set of spills which are likely not to interfere.

This just works for one-word variables for now.  Strings/slices
will be a separate CL.

Change-Id: Ie89eba8cab16bcd41b311c479ec46dd7e64cdb67
Reviewed-on: https://go-review.googlesource.com/16336
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: David Chase <drchase@google.com>
9 years ago[dev.ssa] cmd/compile: introduce storeconst ops
Keith Randall [Wed, 21 Oct 2015 20:13:56 +0000 (13:13 -0700)]
[dev.ssa] cmd/compile: introduce storeconst ops

Introduce opcodes that store a constant value.
AuxInt now needs to hold both the value to be stored and the
constant offset at which to store it.  Introduce a StoreConst
type to help encode/decode these parts to/from an AuxInt.

Change-Id: I1631883abe035cff4b16368683e1eb3d2ccb674d
Reviewed-on: https://go-review.googlesource.com/16170
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
9 years ago[dev.ssa] cmd/compile/internal/ssa: reorder fuse and dse
Todd Neal [Tue, 27 Oct 2015 01:57:53 +0000 (20:57 -0500)]
[dev.ssa] cmd/compile/internal/ssa: reorder fuse and dse

deadstore elimination currently works in a block, fusing before
performing dse eliminates ~1% more stores for make.bash

Change-Id: If5bbddac76bf42616938a8e8e84cb7441fa02f73
Reviewed-on: https://go-review.googlesource.com/16350
Reviewed-by: Keith Randall <khr@golang.org>
9 years ago[dev.ssa] cmd/compile: give GetClosurePtr pointer type
Keith Randall [Tue, 27 Oct 2015 17:15:02 +0000 (10:15 -0700)]
[dev.ssa] cmd/compile: give GetClosurePtr pointer type

If the closure pointer gets spilled, we need to spill it with
pointer type to make stack copy and GC happy.

Change-Id: Ic108748e6b9caecd45522141f02c9422567376e3
Reviewed-on: https://go-review.googlesource.com/16363
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: David Chase <drchase@google.com>
9 years ago[dev.ssa] cmd/compile: better copying
Keith Randall [Thu, 22 Oct 2015 00:18:07 +0000 (17:18 -0700)]
[dev.ssa] cmd/compile: better copying

Replace REP MOVSB with all the copying techniques used by the
old compiler.  Copy in chunks, DUFFCOPY, etc.

Introduces MOVO opcodes and an Int128 type to move around
16 bytes at a time.

Change-Id: I1e73e68ca1d8b3dd58bb4af2f4c9e5d9bf13a502
Reviewed-on: https://go-review.googlesource.com/16174
Reviewed-by: Todd Neal <todd@tneal.org>
Run-TryBot: Keith Randall <khr@golang.org>

9 years ago[dev.ssa] cmd/compile: optimize nil checks
Keith Randall [Sat, 24 Oct 2015 02:12:49 +0000 (19:12 -0700)]
[dev.ssa] cmd/compile: optimize nil checks

Use faulting loads instead of test/jeq to do nil checks.
Fold nil checks into a following load/store if possible.

Makes binaries about 2% smaller.

Change-Id: I54af0f0a93c853f37e34e0ce7e3f01dd2ac87f64
Reviewed-on: https://go-review.googlesource.com/16287
Reviewed-by: David Chase <drchase@google.com>
9 years ago[dev.ssa] cmd/compile: get rid of converts in unsafe.Pointer arithmetic
Keith Randall [Fri, 23 Oct 2015 21:08:50 +0000 (14:08 -0700)]
[dev.ssa] cmd/compile: get rid of converts in unsafe.Pointer arithmetic

unsafe.Pointer->uintptr, add, then uintptr->unsafe.Pointer.
Do the add directly on the pointer type instead.

Change-Id: I5a3a32691d0a000e16975857974ed9a1039c6d28
Reviewed-on: https://go-review.googlesource.com/16281
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: David Chase <drchase@google.com>
9 years ago[dev.ssa] cmd/compile: enhance SSA filtering, add OpConvert
David Chase [Mon, 19 Oct 2015 15:36:07 +0000 (11:36 -0400)]
[dev.ssa] cmd/compile: enhance SSA filtering, add OpConvert

Modified GOSSA{HASH.PKG} environment variable filters to
make it easier to make/run with all SSA for testing.
Disable attempts at SSA for architectures that are not
amd64 (avoid spurious errors/unimplementeds.)

Removed easy out for unimplemented features.

Add convert op for proper liveness in presence of uintptr
to/from unsafe.Pointer conversions.

Tweaked stack sizes to get a pass on windows;
1024 instead 768, was observed to pass at least once.

Change-Id: Ida3800afcda67d529e3b1cf48ca4a3f0fa48b2c5
Reviewed-on: https://go-review.googlesource.com/16201
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: David Chase <drchase@google.com>

9 years ago[dev.ssa] cmd/compile: repair ssa testing build and test
David Chase [Fri, 23 Oct 2015 16:34:03 +0000 (12:34 -0400)]
[dev.ssa] cmd/compile: repair ssa testing build and test

Calls to NewConfig required an extra parameter that
sometimes could not be nil.

Change-Id: I806dd53c045056a0c2d30d641a20fe27fb790539
Reviewed-on: https://go-review.googlesource.com/16272
Reviewed-by: Keith Randall <khr@golang.org>
9 years ago[dev.ssa] cmd/compile: implement reserved registers
Keith Randall [Thu, 22 Oct 2015 20:07:38 +0000 (13:07 -0700)]
[dev.ssa] cmd/compile: implement reserved registers

BP for framepointer experiment
R15 for dynamic linking

Change-Id: I28e48be461d04a4d5c9b013f48fce5c0e58d6a08
Reviewed-on: https://go-review.googlesource.com/16231
Run-TryBot: Todd Neal <todd@tneal.org>
Reviewed-by: Todd Neal <todd@tneal.org>
9 years ago[dev.ssa] cmd/compile: assign unused registers to phi ops
Keith Randall [Mon, 19 Oct 2015 19:24:22 +0000 (12:24 -0700)]
[dev.ssa] cmd/compile: assign unused registers to phi ops

Register phis are better than stack phis.  If we have
unused registers available, use them for phis.

Change-Id: I3045711c65caa1b6d0be29131b87b57466320cc2
Reviewed-on: https://go-review.googlesource.com/16080
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: David Chase <drchase@google.com>
9 years ago[dev.ssa] cmd/compile: don't issue nops for static data
Keith Randall [Tue, 20 Oct 2015 20:56:16 +0000 (13:56 -0700)]
[dev.ssa] cmd/compile: don't issue nops for static data

It confuses live variable analysis to have a bunch of unreachable
no-ops at the end of a function.  Symptom is:
    gc/plive.go:483 panic: interface conversion: interface {} is nil, not *gc.BasicBlock

I don't see any reason why the old compiler needs these no-ops either.
all.bash passes with the equivalent code removed on master.

Change-Id: Ifcd2c3e139aa16314f08aebc9079b2fb7aa60556
Reviewed-on: https://go-review.googlesource.com/16132
Reviewed-by: David Chase <drchase@google.com>
9 years agoMerge "[dev.ssa] Merge remote-tracking branch 'origin/master' into mergebranch" into...
Gerrit Code Review [Tue, 20 Oct 2015 15:46:15 +0000 (15:46 +0000)]
Merge "[dev.ssa] Merge remote-tracking branch 'origin/master' into mergebranch" into dev.ssa

9 years ago[dev.ssa] cmd/compile: getg needs a memory arg
Keith Randall [Tue, 20 Oct 2015 01:54:40 +0000 (18:54 -0700)]
[dev.ssa] cmd/compile: getg needs a memory arg

getg reads from memory, so it should really have a
memory arg.  It is critical in functions which call setg
to make sure getg gets ordered correctly with setg.

Change-Id: Ief4875421f741fc49c07b0e1f065ce2535232341
Reviewed-on: https://go-review.googlesource.com/16100
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: David Chase <drchase@google.com>
Run-TryBot: David Chase <drchase@google.com>

9 years ago[dev.ssa] cmd/compile: don't move mem-using values in tighten pass
Keith Randall [Tue, 20 Oct 2015 01:44:50 +0000 (18:44 -0700)]
[dev.ssa] cmd/compile: don't move mem-using values in tighten pass

It isn't safe, the place where we're moving the value to
might have a different live memory.  Moving will introduce
two simultaneously live memories.

Change-Id: I07e61a6db8ef285088c530dc2e5d5768d27871ff
Reviewed-on: https://go-review.googlesource.com/16099
Reviewed-by: David Chase <drchase@google.com>
Run-TryBot: David Chase <drchase@google.com>

9 years ago[dev.ssa] cmd/compile: Don't rematerialize getg
Keith Randall [Mon, 19 Oct 2015 23:13:58 +0000 (16:13 -0700)]
[dev.ssa] cmd/compile: Don't rematerialize getg

It isn't safe in functions that also call setg.

Change-Id: I76a7bf0401b4b6c8a129c245b15a2d6f06080e94
Reviewed-on: https://go-review.googlesource.com/16095
Reviewed-by: Todd Neal <todd@tneal.org>
9 years ago[dev.ssa] Merge remote-tracking branch 'origin/master' into mergebranch
Keith Randall [Mon, 19 Oct 2015 20:56:55 +0000 (13:56 -0700)]
[dev.ssa] Merge remote-tracking branch 'origin/master' into mergebranch

The only major fixup is that duffzero changed from
8-byte writes to 16-byte writes.

Change-Id: I1762b74ce67a8e4b81c11568027cdb3572f7f87c

9 years agonet: android no longer supports unix/unixgram
David Crawshaw [Mon, 19 Oct 2015 18:41:38 +0000 (14:41 -0400)]
net: android no longer supports unix/unixgram

I cannot find any documentation for this, but these tests no longer run
on the device I have since upgrading to Android L. Presumably it still
works for root, but standard Android programs to not have root access.

Change-Id: I001c8fb5ce22f9ff8d7433f881d0dccbf6ab969d
Reviewed-on: https://go-review.googlesource.com/16056
Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>
9 years ago[dev.ssa] cmd/compile: Rematerialize in regalloc
Keith Randall [Mon, 19 Oct 2015 17:57:03 +0000 (10:57 -0700)]
[dev.ssa] cmd/compile: Rematerialize in regalloc

Rematerialize constants instead of spilling and loading them.
"Constants" includes constant offsets from SP and SB.

Should help somewhat with stack frame sizes.  I'm not sure
exactly how much yet.

Change-Id: I44dbad97aae870cf31cb6e89c92fe4f6a2b9586f
Reviewed-on: https://go-review.googlesource.com/16029
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: David Chase <drchase@google.com>
9 years agoruntime: combine gcResetGState and gcResetMarkState
Austin Clements [Sun, 18 Oct 2015 03:57:53 +0000 (23:57 -0400)]
runtime: combine gcResetGState and gcResetMarkState

These functions are always called together and perform logically
related state resets, so combine them in to just gcResetMarkState.

Fixes #11427.

Change-Id: I06c17ef65f66186494887a767b3993126955b5fe
Reviewed-on: https://go-review.googlesource.com/16041
Reviewed-by: Rick Hudson <rlh@golang.org>
9 years agoruntime: consolidate gcResetGState calls
Austin Clements [Sun, 18 Oct 2015 03:52:49 +0000 (23:52 -0400)]
runtime: consolidate gcResetGState calls

Currently gcResetGState is called by func gcscan_m for concurrent GC
and directly by func gc for STW GC. Simplify this by consolidating
these two calls in to one call by func gc above where it splits for
concurrent and STW GC.

As a consequence, gcResetGState and gcResetMarkState are always called
together, so the next commit will consolidate these.

Change-Id: Ib62d404c7b32b28f7d3080d26ecf3966cbc4aca0
Reviewed-on: https://go-review.googlesource.com/16040
Reviewed-by: Rick Hudson <rlh@golang.org>
9 years agoruntime: remove work.partial queue
Austin Clements [Thu, 15 Oct 2015 19:44:16 +0000 (15:44 -0400)]
runtime: remove work.partial queue

This work queue is no longer used (there are many reads of
work.partial, but the only write is in putpartial, which is never
called).

Fixes #11922.

Change-Id: I08b76c0c02a0867a9cdcb94783e1f7629d44249a
Reviewed-on: https://go-review.googlesource.com/15892
Reviewed-by: Rick Hudson <rlh@golang.org>
9 years agofmt: clarify reflect.Value printing
Nodir Turakulov [Fri, 16 Oct 2015 08:33:28 +0000 (01:33 -0700)]
fmt: clarify reflect.Value printing

fmt docs say:

  If the operand is a reflect.Value, the concrete value it
  holds is printed as if it was the operand.

It implies recursive application of this rule, which is not the case.
Clarify the docs.

Change-Id: I019277c7c6439095bab83e5536aa06403638aa51
Reviewed-on: https://go-review.googlesource.com/15952
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agocmd/go: -buildmode=pie for android/arm
David Crawshaw [Mon, 19 Oct 2015 16:54:38 +0000 (12:54 -0400)]
cmd/go: -buildmode=pie for android/arm

Also make PIE executables the default build mode, as PIE executables
are required as of Android L.

For #10807

Change-Id: I86b7556b9792105cd2531df1b8f3c8f7a8c5d25c
Reviewed-on: https://go-review.googlesource.com/16055
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: David Crawshaw <crawshaw@golang.org>

9 years agocmd/link: PIE executables for android/arm
David Crawshaw [Mon, 19 Oct 2015 16:53:36 +0000 (12:53 -0400)]
cmd/link: PIE executables for android/arm

For #10807

Change-Id: Ied826d06cb622edf6413b6f2cdcc46987ab0b05a
Reviewed-on: https://go-review.googlesource.com/16054
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: David Crawshaw <crawshaw@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agogo/types: check that typed constant shift expressions are representable
Matthew Dempsky [Thu, 15 Oct 2015 16:50:40 +0000 (09:50 -0700)]
go/types: check that typed constant shift expressions are representable

Fixes #12945.

Change-Id: I08b44795fcd7ec59371aea8111f7febead54720b
Reviewed-on: https://go-review.googlesource.com/15900
Reviewed-by: Robert Griesemer <gri@golang.org>
9 years agocmd/cgo: don't print absolute path in comment
David Crawshaw [Mon, 19 Oct 2015 13:11:46 +0000 (09:11 -0400)]
cmd/cgo: don't print absolute path in comment

Change-Id: Ib424e14cfaab35d37ebdd084d41151928bfd645e
Reviewed-on: https://go-review.googlesource.com/16051
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: David Crawshaw <crawshaw@golang.org>

9 years agonet/http: use the correct error variable in serveFile
Brad Fitzpatrick [Mon, 19 Oct 2015 14:45:50 +0000 (14:45 +0000)]
net/http: use the correct error variable in serveFile

It was generating the wrong error message, always defaulting to "500
Internal Server Error", since the err variable used was always nil.

Fixes #12991

Change-Id: I94b0e516409c131ff3b878bcb91e65f0259ff077
Reviewed-on: https://go-review.googlesource.com/16060
Reviewed-by: David Crawshaw <crawshaw@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agoruntime: remove a redundant nil pointer check.
Aaron Jacobs [Mon, 19 Oct 2015 09:52:55 +0000 (20:52 +1100)]
runtime: remove a redundant nil pointer check.

It appears this was made possible by commit 89f185f; before that, g was
not dereferenced above.

Change-Id: I70bc571d924b36351392fd4c13d681e938cfb573
Reviewed-on: https://go-review.googlesource.com/16033
Reviewed-by: Andrew Gerrand <adg@golang.org>
9 years ago[dev.ssa] cmd/compile/internal/ssa: reuse symbols
Todd Neal [Fri, 16 Oct 2015 01:25:32 +0000 (20:25 -0500)]
[dev.ssa] cmd/compile/internal/ssa: reuse symbols

Reuse the ArgSymbol for nodes so that the Aux values will be equal for cse.

Change-Id: Iaae80bd19ff2d3f51b6c9049fd860e04baa6f175
Reviewed-on: https://go-review.googlesource.com/15930
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Todd Neal <todd@tneal.org>

9 years agoruntime: merge proc1.go -> proc.go
Nodir Turakulov [Mon, 19 Oct 2015 00:04:05 +0000 (17:04 -0700)]
runtime: merge proc1.go -> proc.go

from proc1.go to proc.go:
* prepend header comment explaining "Goroutine scheduler"
* insert m0 and g0 var defs after the comment
* append the rest

Updates #12952

Change-Id: I35ee9ae3287675cde0c1b6aeaca0a460393f2354
Reviewed-on: https://go-review.googlesource.com/16024
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agocompress/flate: improve inflate speed by reading more bits at a time
Joe Tsai [Fri, 9 Oct 2015 00:37:07 +0000 (17:37 -0700)]
compress/flate: improve inflate speed by reading more bits at a time

The flate library guarantees that the Reader will never read more
bytes than is necessary. This way, the underlying io.Reader will
be left exactly after the last byte of the DEFLATE stream.
Formats like gzip depend on this behavior being true.

As such, inflate conservatively reads the minimum symbol length in
huffSym leading to many individual calls to moreBits. However, if we
take advantage of the fact that every block *must* end with the EOB
symbol, we can choose to read the length of the EOB symbol.
Since the EOB symbol is also the most rare symbol (occuring exactly
once) in a block, we can hypothesize that it is almost as long as
the max symbol length, allowing huffSym to ask for more bits at the
start of every loop. This increases the probabilty that the Huffman
code is decoded on the first iteration of the outer for-loop.

benchmark                              old MB/s     new MB/s     speedup
BenchmarkDecodeDigitsSpeed1e4-4        51.05        54.31        1.06x
BenchmarkDecodeDigitsSpeed1e5-4        58.86        62.24        1.06x
BenchmarkDecodeDigitsSpeed1e6-4        59.63        63.13        1.06x
BenchmarkDecodeDigitsDefault1e4-4      51.94        54.61        1.05x
BenchmarkDecodeDigitsDefault1e5-4      63.70        69.13        1.09x
BenchmarkDecodeDigitsDefault1e6-4      66.08        71.43        1.08x
BenchmarkDecodeDigitsCompress1e4-4     52.25        54.56        1.04x
BenchmarkDecodeDigitsCompress1e5-4     63.34        68.30        1.08x
BenchmarkDecodeDigitsCompress1e6-4     66.84        70.64        1.06x
BenchmarkDecodeTwainSpeed1e4-4         50.74        53.40        1.05x
BenchmarkDecodeTwainSpeed1e5-4         60.77        67.03        1.10x
BenchmarkDecodeTwainSpeed1e6-4         62.08        69.78        1.12x
BenchmarkDecodeTwainDefault1e4-4       53.45        56.40        1.06x
BenchmarkDecodeTwainDefault1e5-4       73.54        79.05        1.07x
BenchmarkDecodeTwainDefault1e6-4       77.68        83.65        1.08x
BenchmarkDecodeTwainCompress1e4-4      53.21        56.15        1.06x
BenchmarkDecodeTwainCompress1e5-4      73.82        77.76        1.05x
BenchmarkDecodeTwainCompress1e6-4      79.23        83.30        1.05x

Change-Id: Ie194925c827988a380b8c2fdd13b13c4faa5d397
Reviewed-on: https://go-review.googlesource.com/15651
Reviewed-by: Nigel Tao <nigeltao@golang.org>
9 years agoruntime: merge race1.go -> race.go
Nodir Turakulov [Fri, 16 Oct 2015 21:01:45 +0000 (14:01 -0700)]
runtime: merge race1.go -> race.go

* append contents of race1.go to race.go
* delete "Implementation of the race detector API." comment
  from race1.go

Updates #12952

Change-Id: Ibdd9c4dc79a63c3bef69eade9525578063c86c1c
Reviewed-on: https://go-review.googlesource.com/16023
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agoruntime, runtime/cgo: conform to PIC register use rules in ppc64 asm
Michael Hudson-Doyle [Thu, 8 Oct 2015 09:49:39 +0000 (22:49 +1300)]
runtime, runtime/cgo: conform to PIC register use rules in ppc64 asm

PIC code on ppc64le uses R2 as a TOC pointer and when calling a function
through a function pointer must ensure the function pointer is in R12.  These
rules are easy enough to follow unconditionally in our assembly, so do that.

Change-Id: Icfc4e47ae5dfbe15f581cbdd785cdeed6e40bc32
Reviewed-on: https://go-review.googlesource.com/15526
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agoreflect, runtime, runtime/cgo: use ppc64 asm constant for fixed frame size
Michael Hudson-Doyle [Thu, 8 Oct 2015 09:34:29 +0000 (22:34 +1300)]
reflect, runtime, runtime/cgo: use ppc64 asm constant for fixed frame size

Shared libraries on ppc64le will require a larger minimum stack frame (because
the ABI mandates that the TOC pointer is available at 24(R1)). Part 3 of that
is using a #define in the ppc64 assembly to refer to the size of the fixed
part of the stack (finding all these took me about a week!).

Change-Id: I50f22fe1c47af1ec59da1bd7ea8f84a4750df9b7
Reviewed-on: https://go-review.googlesource.com/15525
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agocmd/compile, cmd/internal/obj: centralize knowledge of size of fixed part of stack
Michael Hudson-Doyle [Thu, 8 Oct 2015 09:13:44 +0000 (22:13 +1300)]
cmd/compile, cmd/internal/obj: centralize knowledge of size of fixed part of stack

Shared libraries on ppc64le will require a larger minimum stack frame (because
the ABI mandates that the TOC pointer is available at 24(R1)). Part 2a of
preparing for that is to have all bits of arch-independent and ppc64-specific
codegen that need to know call a function to find out.

Change-Id: I55899f73037e92227813c491049a3bd6f30bd41f
Reviewed-on: https://go-review.googlesource.com/15524
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agoruntime: add a constant for the smallest possible stack frame
Michael Hudson-Doyle [Thu, 8 Oct 2015 08:52:03 +0000 (21:52 +1300)]
runtime: add a constant for the smallest possible stack frame

Shared libraries on ppc64le will require a larger minimum stack frame (because
the ABI mandates that the TOC pointer is available at 24(R1)). So to prepare
for this, make a constant for the fixed part of a stack and use that where
necessary.

Change-Id: I447949f4d725003bb82e7d2cf7991c1bca5aa887
Reviewed-on: https://go-review.googlesource.com/15523
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Michael Hudson-Doyle <michael.hudson@canonical.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocmd/internal/obj, runtime: add NOFRAME flag to suppress stack frame set up on ppc64x
Michael Hudson-Doyle [Thu, 8 Oct 2015 23:44:27 +0000 (12:44 +1300)]
cmd/internal/obj, runtime: add NOFRAME flag to suppress stack frame set up on ppc64x

Replace the confusing game where a frame size of $-8 would suppress the
implicit setting up of a stack frame with a nice explicit flag.

The code to set up the function prologue is still a little confusing but better
than it was.

Change-Id: I1d49278ff42c6bc734ebfb079998b32bc53f8d9a
Reviewed-on: https://go-review.googlesource.com/15670
Reviewed-by: Minux Ma <minux@golang.org>
9 years agomime: Remove an allocation in word decoding.
Justin Nuß [Wed, 9 Sep 2015 19:10:43 +0000 (21:10 +0200)]
mime: Remove an allocation in word decoding.

This fixes a TODO in (*WordDecoder).Decode by replacing a call to strings.Split with
simple prefix/suffix checking and some custom slicing.

Benchmark results:

benchmark                    old ns/op     new ns/op     delta
BenchmarkQEncodeWord-8       740           693           -6.35%
BenchmarkQDecodeWord-8       1291          727           -43.69%
BenchmarkQDecodeHeader-8     1194          767           -35.76%

benchmark                    old allocs     new allocs     delta
BenchmarkQEncodeWord-8       1              1              +0.00%
BenchmarkQDecodeWord-8       3              2              -33.33%
BenchmarkQDecodeHeader-8     3              2              -33.33%

benchmark                    old bytes     new bytes     delta
BenchmarkQEncodeWord-8       48            48            +0.00%
BenchmarkQDecodeWord-8       128           48            -62.50%
BenchmarkQDecodeHeader-8     128           48            -62.50%

Change-Id: I2d6844c75ec2e2b79be2e49b7fc4ca320b7e84e5
Reviewed-on: https://go-review.googlesource.com/14432
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years ago[dev.ssa] cmd/compile: Reuse stack slots for spill locations
Keith Randall [Sun, 18 Oct 2015 04:14:56 +0000 (21:14 -0700)]
[dev.ssa] cmd/compile: Reuse stack slots for spill locations

For each type, maintain a list of stack slots used to spill
SSA values to the stack.  Reuse those stack slots for noninterfering
spills.

Lowers frame sizes.  As an example, runtime.mSpan_Sweep goes from
584 bytes to 392 bytes.  heapBitsSetType goes from 576 bytes to 152 bytes.

Change-Id: I0e9afe80c2fd84aff9eb368318685de293c363d0
Reviewed-on: https://go-review.googlesource.com/16022
Reviewed-by: David Chase <drchase@google.com>
9 years ago[dev.ssa] cmd/compile: fill remaining SSA gaps
David Chase [Fri, 9 Oct 2015 20:48:30 +0000 (16:48 -0400)]
[dev.ssa] cmd/compile: fill remaining SSA gaps

Changed racewalk/race detector to use FP in a more
sensible way.

Relaxed checks for CONVNOP when race detecting.

Modified tighten to ensure that GetClosurePtr cannot float
out of entry block (turns out this cannot be relaxed, DX is
sometimes stomped by other code accompanying race detection).

Added case for addr(CONVNOP)

Modified addr to take "bounded" flag to suppress nilchecks
where it is set (usually, by race detector).

Cannot leave unimplemented-complainer enabled because it
turns out we are optimistically running SSA on every platform.

Change-Id: Ife021654ee4065b3ffac62326d09b4b317b9f2e0
Reviewed-on: https://go-review.googlesource.com/15710
Reviewed-by: Keith Randall <khr@golang.org>
9 years agoruntime: merge stack{1,2}.go -> stack.go
Nodir Turakulov [Sat, 17 Oct 2015 01:45:30 +0000 (18:45 -0700)]
runtime: merge stack{1,2}.go -> stack.go

* rename stack1.go -> stack.go
* prepend contents of stack2.go to stack.go

Updates #12952

Change-Id: I60d409af37162a5a7596c678dfebc2cea89564ff
Reviewed-on: https://go-review.googlesource.com/16008
Reviewed-by: Austin Clements <austin@google.com>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocmd/link: always disable lazy PLT resolution when dynamically linking Go
Michael Hudson-Doyle [Thu, 15 Oct 2015 08:48:11 +0000 (21:48 +1300)]
cmd/link: always disable lazy PLT resolution when dynamically linking Go

Go cannot allow lazy PLT resolution when calling between Go functions because
the lazy resolution can use more stack than is available. Lazy resolution is
disabled by passing -z now to the system linker, but unfortunately was only
passed when linking to a Go shared library. That sounds fine, but the shared
library containing the runtime is not linked to any other Go shared library but
calls main.init and main.main via a PLT, and before this fix this did use lazy
resolution. (For some reason this never caused a problem on intel, but it
breaks on ppc64le). Fortunately the fix is very simple: always pass -z now to
the system linker when dynamically linking Go.

Change-Id: I7806d40aac80dcd1e56b95864d1cfeb1c42614e2
Reviewed-on: https://go-review.googlesource.com/15870
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agonet/http: Client support for Expect: 100-continue
Yasuharu Goto [Thu, 14 May 2015 15:44:34 +0000 (00:44 +0900)]
net/http: Client support for Expect: 100-continue

Current http client doesn't support Expect: 100-continue request
header(RFC2616-8/RFC7231-5.1.1). So even if the client have the header,
the head of the request body is consumed prematurely.

Those are my intentions to avoid premature consuming body in this change.
- If http.Request header contains body and Expect: 100-continue
  header, it blocks sending body until it gets the first response.
- If the first status code to the request were 100, the request
  starts sending body. Otherwise, sending body will be cancelled.
- Tranport.ExpectContinueTimeout specifies the amount of the time to
  wait for the first response.

Fixes #3665

Change-Id: I4c04f7d88573b08cabd146c4e822061764a7cd1f
Reviewed-on: https://go-review.googlesource.com/10091
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agoruntime: remove some unnecessary unsafe code in mfixalloc
Matthew Dempsky [Fri, 16 Oct 2015 23:21:26 +0000 (16:21 -0700)]
runtime: remove some unnecessary unsafe code in mfixalloc

Change-Id: Ie9ea4af4315a4d0eb69d0569726bb3eca2b397af
Reviewed-on: https://go-review.googlesource.com/16005
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agonet: unblock plan9 TCP Read calls after socket close
Brad Fitzpatrick [Fri, 16 Oct 2015 16:28:42 +0000 (16:28 +0000)]
net: unblock plan9 TCP Read calls after socket close

Fixes #7782
Fixes #9554
Updates #7237 (original metabug, before we switched to specific bugs)
Updates #11932 (plan9 still doesn't have net I/O deadline support)

Change-Id: I96f311b88b1501d884ebc008fd31ad2cf1e16d75
Reviewed-on: https://go-review.googlesource.com/15941
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: David du Colombier <0intro@gmail.com>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agonet: use IndexByte implementation from runtime package
Ilya Tocar [Thu, 15 Oct 2015 15:59:01 +0000 (18:59 +0300)]
net: use IndexByte implementation from runtime package

In net/parse.go we reimplement bytes.IndexByte and strings.IndexByte,
However those are implemented in runtime/$GOARCH_asm.s.
Using versions from runtime should provide performance advantage,
and keep the same code together.

Change-Id: I6212184bdf6aa1f2c03ce26d4b63f5b379d8ed0c
Reviewed-on: https://go-review.googlesource.com/15953
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agoruntime: merge panic1.go into panic.go
Nodir Turakulov [Fri, 16 Oct 2015 02:00:12 +0000 (19:00 -0700)]
runtime: merge panic1.go into panic.go

A TODO to merge is removed from panic1.go.
The rest is appended to panic.go

Updates #12952

Change-Id: Ied4382a455abc20bc2938e34d031802e6b4baf8b
Reviewed-on: https://go-review.googlesource.com/15905
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>

9 years agoruntime: rename print1.go -> print.go
Nodir Turakulov [Fri, 16 Oct 2015 02:30:02 +0000 (19:30 -0700)]
runtime: rename print1.go -> print.go

It seems that it was called print1.go mistakenly: print.go was deleted
in the same commit:
https://go.googlesource.com/go/+/597b266eafe7d63e9be8da1c1b4813bd2998a11c

Updates #12952

Change-Id: I371e59d6cebc8824857df3f3ee89101147dfffc0
Reviewed-on: https://go-review.googlesource.com/15950
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>

9 years agoruntime: merge string1.go into string.go
Nodir Turakulov [Fri, 16 Oct 2015 06:34:56 +0000 (23:34 -0700)]
runtime: merge string1.go into string.go

string1.go contents are appended to string.go as is

Updates #12952

Change-Id: I30083ba7fdd362d4421e964a494c76ca865bedc2
Reviewed-on: https://go-review.googlesource.com/15951
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agodatabase/sql: avoid deadlock waiting for connections
Chris Hines [Mon, 14 Sep 2015 07:44:56 +0000 (03:44 -0400)]
database/sql: avoid deadlock waiting for connections

Previously with db.maxOpen > 0, db.maxOpen+n failed connection attempts
started concurrently could result in a deadlock. DB.conn and
DB.openNewConnection did not trigger the DB.connectionOpener go routine
after a failed connection attempt. This omission could leave go routines
waiting for DB.connectionOpener forever.

In addition the logic to track the state of the pool was inconsistent.
db.numOpen was sometimes incremented optimistically and sometimes not.
This change harmonizes the logic and eliminates the db.pendingOpens
variable, making the logic easier to understand and maintain.

Fixes #10886

Change-Id: I983c4921a3dacfbd531c3d7f8d2da8a592e9922a
Reviewed-on: https://go-review.googlesource.com/14547
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agosyscall: fix nacl listener to not accept connections once closed
Brad Fitzpatrick [Fri, 16 Oct 2015 04:41:34 +0000 (04:41 +0000)]
syscall: fix nacl listener to not accept connections once closed

Change-Id: I3ba6e97089ef6e69ba31dfb632df465859906a74
Reviewed-on: https://go-review.googlesource.com/15940
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Minux Ma <minux@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agoruntime, runtime/debug: access unexported runtime functions with //go:linkname, not...
Michael Hudson-Doyle [Fri, 16 Oct 2015 07:19:14 +0000 (20:19 +1300)]
runtime, runtime/debug: access unexported runtime functions with //go:linkname, not assembly stubs

Change-Id: I88f80f5914d6e4c179f3d28aa59fc29b7ef0cc66
Reviewed-on: https://go-review.googlesource.com/15960
Reviewed-by: Minux Ma <minux@golang.org>
Run-TryBot: Minux Ma <minux@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agoruntime, os/signal: use //go:linkname instead of assembly stubs to get access to...
Michael Hudson-Doyle [Thu, 15 Oct 2015 20:56:07 +0000 (09:56 +1300)]
runtime, os/signal: use //go:linkname instead of assembly stubs to get access to runtime functions

os/signal depends on a few unexported runtime functions. This removes the
assembly stubs it used to get access to these in favour of using
//go:linkname in runtime to make the functions accessible to os/signal.

This is motivated by ppc64le shared libraries, where you cannot BR to a symbol
defined in a shared library (only BL), but it seems like an improvment anyway.

Change-Id: I09361203ce38070bd3f132f6dc5ac212f2dc6f58
Reviewed-on: https://go-review.googlesource.com/15871
Run-TryBot: Michael Hudson-Doyle <michael.hudson@canonical.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Minux Ma <minux@golang.org>
Reviewed-by: Dave Cheney <dave@cheney.net>
9 years agoruntime: use unsafe.Pointer(x) instead of (unsafe.Pointer)(x)
Matthew Dempsky [Thu, 15 Oct 2015 21:33:50 +0000 (14:33 -0700)]
runtime: use unsafe.Pointer(x) instead of (unsafe.Pointer)(x)

This isn't C anymore.  No binary change to pkg/linux_amd64/runtime.a.

Change-Id: I24d66b0f5ac888f432b874aac684b1395e7c8345
Reviewed-on: https://go-review.googlesource.com/15903
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agocmd/compile/internal/gc: reject invalid unsafe.Sizeof([0]byte{}[0])
Matthew Dempsky [Thu, 15 Oct 2015 18:29:06 +0000 (11:29 -0700)]
cmd/compile/internal/gc: reject invalid unsafe.Sizeof([0]byte{}[0])

Apply static bounds checking logic during type checking even to
zero-element arrays, but skip synthesized OINDEX nodes that the
compiler has asserted are within bounds (such as the ones generated
while desugaring ORANGE nodes).  This matches the logic in walkexpr
that also skips static bounds checking when Bounded is true.

Passes toolstash/buildall.

Fixes #12944.

Change-Id: I14ba03d71c002bf969d69783bec8d1a8e10e7d75
Reviewed-on: https://go-review.googlesource.com/15902
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agocmd/compile/internal/gc: refactor range/memclr optimization
Matthew Dempsky [Thu, 15 Oct 2015 18:08:09 +0000 (11:08 -0700)]
cmd/compile/internal/gc: refactor range/memclr optimization

No functional change and passes toolstash/buildall, but eliminates a
13-deep nesting of if statements.

Change-Id: I32e63dcf358c6eb521935f4ee07fbe749278e5ef
Reviewed-on: https://go-review.googlesource.com/15901
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agonet/http/cgi: optimize internal function removeLeadingDuplicates a bit
Brad Fitzpatrick [Wed, 14 Oct 2015 15:12:40 +0000 (15:12 +0000)]
net/http/cgi: optimize internal function removeLeadingDuplicates a bit

Change-Id: I0255f24f5c5925ea4daa28a28d23606df35d4373
Reviewed-on: https://go-review.googlesource.com/15824
Reviewed-by: Andrew Gerrand <adg@golang.org>
9 years agotext/template: resolve non-empty interface
Nodir Turakulov [Wed, 14 Oct 2015 02:09:32 +0000 (19:09 -0700)]
text/template: resolve non-empty interface

Read what a non-empty interface points to.

The deleted lines were added in https://codereview.appspot.com/4810060/,
which attempted to break an infinite loop. That was a long time ago.
If I just delete these lines with current codebase, the test "bug1"
(added in that CL) does not fail.

All new tests fail without this fix.

Fixes #12924

Change-Id: I9370ca44facd6af3019850aa065b936e5a482d37
Reviewed-on: https://go-review.googlesource.com/15809
Reviewed-by: Andrew Gerrand <adg@golang.org>
9 years agomime: limit UTF-8 encoded-word length to 75 characters
Alexandre Cesaro [Thu, 24 Sep 2015 21:45:13 +0000 (23:45 +0200)]
mime: limit UTF-8 encoded-word length to 75 characters

As specified by RFC 2047 section 2, encoded-words may not be more than
75 characters long.

We only enforce this rule when the charset is UTF-8, since multi-bytes
characters must not be split accross encoded-words (see section 5.3).

Fixes #12300

Change-Id: I72a43fc3fe6ddeb3dab54dcdce0837d7ebf658f0
Reviewed-on: https://go-review.googlesource.com/14957
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agonet/http/httputil: clarify where DumpRequest(Out)? can be used.
Aaron Jacobs [Wed, 14 Oct 2015 23:25:56 +0000 (10:25 +1100)]
net/http/httputil: clarify where DumpRequest(Out)? can be used.

The clues to this were already there, but as a user I was still unsure.
Make this more explicit.

Change-Id: I68564f3498dcd4897772a303588f03a6b65f111d
Reviewed-on: https://go-review.googlesource.com/15172
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agoencoding/json: simplify encodeState.{string, stringBytes}
Nodir Turakulov [Wed, 14 Oct 2015 22:18:10 +0000 (15:18 -0700)]
encoding/json: simplify encodeState.{string, stringBytes}

As correctly mentioned in #11883, encodeState.string and
encodeState.stringBytes never return an error.
This CL removes the error from the function signatures and somewhat
simplifies call sites.

Fixes #11883

Change-Id: I1d1853d09631c545b68b5eea86ff7daa2e0ca10b
Reviewed-on: https://go-review.googlesource.com/15836
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agonet/http: enable automatic HTTP/2 if TLSNextProto is nil
Brad Fitzpatrick [Wed, 14 Oct 2015 20:41:36 +0000 (20:41 +0000)]
net/http: enable automatic HTTP/2 if TLSNextProto is nil

This enables HTTP/2 by default (for https only) if the user didn't
configure anything in their NPN/ALPN map. If they're using SPDY or an
alternate http2 or a newer http2 from x/net/http2, we do nothing
and don't use the standard library's vendored copy of x/net/http2.

Upstream remains golang.org/x/net/http2.

Update #6891

Change-Id: I69a8957a021a00ac353f9d7fdb9a40a5b69f2199
Reviewed-on: https://go-review.googlesource.com/15828
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Andrew Gerrand <adg@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agovendor: import golang.org/x/net/http2
Brad Fitzpatrick [Wed, 14 Oct 2015 03:37:10 +0000 (03:37 +0000)]
vendor: import golang.org/x/net/http2

golang.org/x/net/http2 becomes net/http/h2_bundle.go (using adonovan's
x/tools/cmd/bundle tool), becoming a hidden part of the net/http package.

golang.org/x/net/http2/hpack becomes vendor/golang.org/x/net/http2/hpack.

At git rev 7331ef52 (https://go-review.googlesource.com/15821)

Change-Id: Ia6683e6f91a481b11a778638bf65b6a338744eea
Reviewed-on: https://go-review.googlesource.com/15822
Reviewed-by: Andrew Gerrand <adg@golang.org>
9 years agonet/http: don't hang if RemoteAddr() blocks
David Glasser [Wed, 14 Oct 2015 21:25:00 +0000 (14:25 -0700)]
net/http: don't hang if RemoteAddr() blocks

The PROXY protocol is supported by several proxy servers such as haproxy
and Amazon ELB.  This protocol allows services running behind a proxy to
learn the remote address of the actual client connecting to the proxy,
by including a single textual line at the beginning of the TCP
connection.
http://www.haproxy.org/download/1.5/doc/proxy-protocol.txt

There are several Go libraries for this protocol (such as
https://github.com/armon/go-proxyproto), which operate by wrapping a
net.Conn with an implementation whose RemoteAddr method reads the
protocol line before returning. This means that RemoteAddr is a blocking
call.

Before this change, http.Serve called RemoteAddr from the main Accepting
goroutine, not from the per-connection goroutine. This meant that it
would not Accept another connection until RemoteAddr returned, which is
not appropriate if RemoteAddr needs to do a blocking read from the
socket first.

Fixes #12943.

Change-Id: I1a242169e6e4aafd118b794e7c8ac45d0d573421
Reviewed-on: https://go-review.googlesource.com/15835
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocompress/gzip: fix go vet warning
Joe Tsai [Tue, 13 Oct 2015 20:19:09 +0000 (13:19 -0700)]
compress/gzip: fix go vet warning

Change-Id: I2ac6b178e666e34e028001078faf62ce71e12e9d
Reviewed-on: https://go-review.googlesource.com/15834
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years ago[dev.ssa] cmd/compile: allow rewrite rules to specify a target block
Keith Randall [Tue, 13 Oct 2015 18:08:08 +0000 (11:08 -0700)]
[dev.ssa] cmd/compile: allow rewrite rules to specify a target block

Some rewrite rules need to make sure the rewrite target ends up
in a specific block.  For example:

(MOVBQSX (MOVBload [off] {sym} ptr mem)) ->
   @v.Args[0].Block (MOVBQSXload <v.Type> [off] {sym} ptr mem)

The MOVBQSXload op needs to be in the same block as the MOVBload
(to ensure exactly one memory is live at basic block boundaries).

Change-Id: Ibe49a4183ca91f6c859cba8135927f01d176e064
Reviewed-on: https://go-review.googlesource.com/15804
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
9 years agonet/http/cgi: make provided Env override even system env vars
Steven Hartland [Fri, 25 Sep 2015 08:51:50 +0000 (08:51 +0000)]
net/http/cgi: make provided Env override even system env vars

Allow all CGI environment settings from the inherited set and default
inherited set to be overridden including PATH by Env.

Change-Id: Ief8d33247b879fa87a8bfd6416d4813116db98de
Reviewed-on: https://go-review.googlesource.com/14959
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>

9 years agoruntime: Reduce testing for fastlog2 implementation
Raul Silvera [Wed, 14 Oct 2015 02:49:40 +0000 (19:49 -0700)]
runtime: Reduce testing for fastlog2 implementation

The current fastlog2 testing checks all 64M values in the domain of
interest, which is too much for platforms with no native floating point.

Reduce testing under testing.Short() to speed up builds for those platforms.

Related to #12620

Change-Id: Ie5dcd408724ba91c3b3fcf9ba0dddedb34706cd1
Reviewed-on: https://go-review.googlesource.com/15830
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Joel Sing <jsing@google.com>
Reviewed-by: Minux Ma <minux@golang.org>
Run-TryBot: Minux Ma <minux@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agoencoding/csv: indicate package of EOF in docs
Katrina Owen [Tue, 13 Oct 2015 22:42:20 +0000 (16:42 -0600)]
encoding/csv: indicate package of EOF in docs

The documentation listing err == EOF can be confusing to newcomers
to the language who are looking for the relevant documentation for
that error.

Change-Id: I301885950d0e1d0fbdf3a1892fca86eac7a0c616
Reviewed-on: https://go-review.googlesource.com/15806
Reviewed-by: Andrew Gerrand <adg@golang.org>
9 years agocmd/internal/obj: move empty field to the top
Shenghou Ma [Fri, 9 Oct 2015 00:24:17 +0000 (20:24 -0400)]
cmd/internal/obj: move empty field to the top

Due to #9401, trailing empty fields will occupy at least 1 byte
of space.

Fixes #12884.

Change-Id: I838d3f1a73637e526f5a6dbc348981227d5bb2fd
Reviewed-on: https://go-review.googlesource.com/15660
Run-TryBot: Minux Ma <minux@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agomisc/cgo/testcshared: use -pie for android-L.
Hyang-Ah Hana Kim [Tue, 13 Oct 2015 17:58:11 +0000 (13:58 -0400)]
misc/cgo/testcshared: use -pie for android-L.

Also, handle the case where 'read' returns EINVAL instead of EBADF
when the descriptor is not ready. (android 4.4.4/cyanogenmod, nexus7)

Change-Id: I56c5949d27303d44a4fd0de38951b85e20cef167
Reviewed-on: https://go-review.googlesource.com/15810
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agomisc/cgo/test: fix go vet warnings
Ian Lance Taylor [Tue, 13 Oct 2015 17:05:13 +0000 (10:05 -0700)]
misc/cgo/test: fix go vet warnings

Fixes these warnings from go vet:
buildid_linux.go:25: no formatting directive in Fatalf call
callback.go:180: arg pc[i] for printf verb %p of wrong type: uintptr
env.go:34: possible misuse of unsafe.Pointer
issue7665.go:22: possible misuse of unsafe.Pointer

Change-Id: I83811b9c10c617139713a626b4a34ab05564d4fe
Reviewed-on: https://go-review.googlesource.com/15802
Reviewed-by: David Crawshaw <crawshaw@golang.org>
9 years agonet/http: ignore Transfer-Encoding for HTTP/1.0 responses
Brad Fitzpatrick [Tue, 13 Oct 2015 00:29:44 +0000 (00:29 +0000)]
net/http: ignore Transfer-Encoding for HTTP/1.0 responses

Fixes #12785

Change-Id: Iae4383889298c6a78b1ba41bd2cda70b0758fcba
Reviewed-on: https://go-review.googlesource.com/15737
Reviewed-by: Andrew Gerrand <adg@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agoruntime: remove _Kind constants
Ian Lance Taylor [Mon, 12 Oct 2015 23:01:51 +0000 (16:01 -0700)]
runtime: remove _Kind constants

The duplication of _Kind and kind constants is a legacy of the
conversion from C.

Change-Id: I368b35a41f215cf91ac4b09dac59699edb414a0e
Reviewed-on: https://go-review.googlesource.com/15800
Reviewed-by: Minux Ma <minux@golang.org>
Run-TryBot: Minux Ma <minux@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agocmd/go: always log dynamic import errors
Burcu Dogan [Mon, 12 Oct 2015 08:02:51 +0000 (01:02 -0700)]
cmd/go: always log dynamic import errors

There is no easy way to understand what user intent was and whether
they wanted to use a dynamic import or not.

If we skip logging such errors, it breaks common use cases such as
https://golang.org/issue/12810.

It's a better approach to expose the underlying mechanism and
be more verbose with the error messages.

Fixes #12810.

Change-Id: I7e922c9e848382690d9d9b006d7046e6cf93223b
Reviewed-on: https://go-review.googlesource.com/15756
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Andrew Gerrand <adg@golang.org>
9 years agocmd/go: don't accept a relative path as GOBIN
Burcu Dogan [Mon, 12 Oct 2015 15:56:05 +0000 (08:56 -0700)]
cmd/go: don't accept a relative path as GOBIN

Fixes #12907.

Change-Id: I5925852fe6962d4ec7dbb3ea5323e8ddfaf9d576
Reviewed-on: https://go-review.googlesource.com/15755
Reviewed-by: Andrew Gerrand <adg@golang.org>
Run-TryBot: Andrew Gerrand <adg@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agoapi: update next.txt
Brad Fitzpatrick [Mon, 12 Oct 2015 22:40:10 +0000 (22:40 +0000)]
api: update next.txt

Change-Id: I536d55f784bb042a2dee94b2878aff42bc182eed
Reviewed-on: https://go-review.googlesource.com/15734
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agonet/http: export the "new" error codes from RFC 6585
Brad Fitzpatrick [Mon, 12 Oct 2015 20:41:40 +0000 (20:41 +0000)]
net/http: export the "new" error codes from RFC 6585

These were proposed in the RFC over three years ago, then proposed to
be added to Go in https://codereview.appspot.com/7678043/ 2 years and
7 months ago, and the spec hasn't been updated or retracted the whole
time.

Time to export them.

Of note, HTTP/2 uses code 431 (Request Header Fields Too Large).

Updates #12843

Change-Id: I78c2fed5fab9540a98e845ace73f21c430a48809
Reviewed-on: https://go-review.googlesource.com/15732
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agonet/textproto: properly trim continued lines in MIME headers
Didier Spezia [Sat, 10 Oct 2015 16:53:09 +0000 (16:53 +0000)]
net/textproto: properly trim continued lines in MIME headers

A MIME header can include values defined on several lines.
Only the first line of each value was trimmed.

Make sure all the lines are trimmed before being aggregated.

Fixes #11204

Change-Id: Id92f384044bc6c4ca836e5dba2081fe82c82dc85
Reviewed-on: https://go-review.googlesource.com/15683
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agoimage: add NYCbCrA types.
Nigel Tao [Sat, 10 Oct 2015 00:25:47 +0000 (11:25 +1100)]
image: add NYCbCrA types.

Fixes #12722

Change-Id: I6a630d8b072ef2b1c63de941743148f8c96b8e5f
Reviewed-on: https://go-review.googlesource.com/15671
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agoos: make IsPermission, IsExist, and IsNotExist unwrap SyscallError too
Brad Fitzpatrick [Sat, 10 Oct 2015 04:15:48 +0000 (04:15 +0000)]
os: make IsPermission, IsExist, and IsNotExist unwrap SyscallError too

Fixes #12674

Change-Id: I82f53026dd2fc27bd7999d43c27932d683d92af6
Reviewed-on: https://go-review.googlesource.com/15730
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agodoc: go1.6.txt: *url.Error implements net.Error
Dave Cheney [Sat, 10 Oct 2015 11:35:13 +0000 (22:35 +1100)]
doc: go1.6.txt: *url.Error implements net.Error

Change-Id: I34c8ada1f3c5d401944483df424011fa2ae9fc3d
Reviewed-on: https://go-review.googlesource.com/15673
Reviewed-by: Dave Cheney <dave@cheney.net>
9 years agonet/url: make *url.Error implement net.Error
Dave Cheney [Sat, 10 Oct 2015 03:21:42 +0000 (14:21 +1100)]
net/url: make *url.Error implement net.Error

Fixes #12866

net/http.Client returns some errors wrapped in a *url.Error. To avoid
the requirement to unwrap these errors to determine if the cause was
temporary or a timeout, make *url.Error implement net.Error directly.

Change-Id: I1ba84ecc7ad5147a40f056ff1254e60290152408
Reviewed-on: https://go-review.googlesource.com/15672
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>

9 years agocompress/flate: deprecate ReadError and WriteError
Joe Tsai [Tue, 22 Sep 2015 08:33:11 +0000 (01:33 -0700)]
compress/flate: deprecate ReadError and WriteError

A vast majority of the time, ReadError isn't even returned during
IO operations. Instead, an unwrapped error will be returned because
of the ReadByte call on L705. Because DEFLATE streams are primarily
compressed and require byte for byte Huffman decoding, most of the
data read from a data stream will go through ReadByte.

Although this is technically an API change, any user reliant on
this error would not have worked properly anyways due to the fact
that most IO error are not wrapped. We might as well deprecate
ReadError. It is useless and actually makes clients that do
depend on catching IO errors more difficult.

Fixes #11856
Fixes #12724

Change-Id: Ib5fec5ae215e977c4e85de5701ce6a473d400af8
Reviewed-on: https://go-review.googlesource.com/14834
Reviewed-by: Nigel Tao <nigeltao@golang.org>
9 years agocmd/link: remove -W option
Ian Lance Taylor [Fri, 9 Oct 2015 20:02:25 +0000 (13:02 -0700)]
cmd/link: remove -W option

The -W option has not worked since Go 1.3.  It is not documented.  When
it did work, it generated useful output, but it was for human viewing;
there was no reason to write a script that passes the -W option, so it's
unlikely that anybody is using it today.

Change-Id: I4769f1ffd308a48324a866592eb7fd79a4cdee54
Reviewed-on: https://go-review.googlesource.com/15701
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>