Rémy Oudompheng [Wed, 2 Jan 2013 21:55:23 +0000 (22:55 +0100)]
cmd/dist, cmd/8g: implement GO386=387/sse to choose FPU flavour.
A new environment variable GO386 is introduced to choose between
code generation targeting 387 or SSE2. No auto-detection is
performed and the setting defaults to 387 to preserve previous
behaviour.
The patch is a reorganization of CL6549052 by rsc.
Rémy Oudompheng [Wed, 2 Jan 2013 20:42:26 +0000 (21:42 +0100)]
cmd/gc: mark wrapper methods for unnamed types as DUPOK.
Unnamed types like structs with embedded fields can have methods.
These methods are generated on-the-fly by the compiler and
it may happen for identical types in different packages.
The linker must accept these multiple definitions.
Adam Langley [Sat, 29 Dec 2012 00:11:37 +0000 (19:11 -0500)]
crypto/rsa: ensure that RSA keys use the full number of bits.
While half of all numbers don't have their most-significant bit set,
this is becoming increasingly impermissible for RSA moduli. In an
attempt to exclude weak keys, several bits of software either do, or
will, enforce that RSA moduli are >= 1024-bits.
However, Go often generates 1023-bit RSA moduli which this software
would then reject.
This change causes crypto/rsa to regenerate the primes in the event
that the result is shorter than requested.
It also alters crypto/rand in order to remove the performance impact
of this:
The most important change to crypto/rand is that it will now set the
top two bits in a generated prime (OpenSSL does the same thing).
Multiplying two n/2 bit numbers, where each have the top two bits set,
will always result in an n-bit product. (The effectively makes the
crypto/rsa change moot, but that seems too fragile to depend on.)
Also this change adds code to crypto/rand to rapidly eliminate some
obviously composite numbers and reduce the number of Miller-Rabin
tests needed to generate a prime.
Robert Griesemer [Fri, 28 Dec 2012 18:40:36 +0000 (10:40 -0800)]
exp/types: resolve composite literal keys
The parser/resolver cannot accurately resolve
composite literal keys that are identifiers;
it needs type information.
Instead, try to resolve them but leave final
judgement to the type checker.
Jan Ziak [Thu, 27 Dec 2012 18:35:04 +0000 (02:35 +0800)]
reflect: declare slice as *[]unsafe.Pointer instead of *[]byte
The new garbage collector (CL 6114046) may find the fake *[]byte value
and interpret its contents as bytes rather than as potential pointers.
This may lead the garbage collector to free memory blocks that
shouldn't be freed.
Akshat Kumar [Thu, 27 Dec 2012 00:45:03 +0000 (11:45 +1100)]
log/syslog: add stub for Plan 9
Proper local system log semantics still need to be
created for Plan 9. In the meantime, the test suite
(viz., exp/gotype) expects there to be some Go
source for each import path. Thus, here is a stub,
equivalent to syslog_windows, for this purpose.
Robert Griesemer [Wed, 26 Dec 2012 20:48:26 +0000 (12:48 -0800)]
exp/types: configurable types.Check API
- added Context type for configuration of type checker
- type check all function and method bodies
- (partial) fixes to shift hinting (still not complete)
- revamped test harness - does not rely on specific position
representation anymore, just a standard (compiler) error
message
- lots of bug fixes
Marcel van Lohuizen [Mon, 24 Dec 2012 15:53:25 +0000 (16:53 +0100)]
exp/norm: changed API of Iter.
Motivations:
- Simpler UI. Previous API proved a bit awkward for practical purposes.
- Iter is often used in cases where one want to be able to bail out early.
The old implementaton had too much look-ahead to be efficient.
Disadvantages:
- ASCII performance is bad. This is unavoidable for tiny iterations.
Example is included to show how to work around this.
Description:
Iter now iterates per boundary/segment. It returns a slice of bytes that
either points to the input bytes, the internal decomposition strings,
or the small internal buffer that each iterator has. In many cases, copying
bytes is avoided.
The method Seek was added to support jumping around the input without
having to reinitialize.
Details:
- Table adjustments: some decompositions exist of multiple segments.
Decompositions that are of this type are now marked so that Iter can
handle them separately.
- The old iterator had a different next function for different normal forms
that was assigned to a function pointer called by Next.
The new iterator uses this mechanism to switch between different modes
for handling different type of input as well. This greatly improves
performance for Hangul and ASCII. It is also used for multi-segment
decompositions.
- input is now a struct of sting and []byte, instead of an interface.
This simplifies optimizing the ASCII case.
Marcel van Lohuizen [Mon, 24 Dec 2012 15:42:29 +0000 (16:42 +0100)]
exp/locale/collate: include composed characters into the table. This eliminates
the need to decompose characters for the majority of cases. This considerably
speeds up collation while increasing the table size minimally.
To detect non-normalized strings, rather than relying on exp/norm, the table
now includes CCC information. The inclusion of this information does not
increase table size.
DETAILS
- Raw collation elements are now a struct that includes the CCC, rather
than a slice of ints.
- Builder now ensures that NFD and NFC counterparts are included in the table.
This also fixes a bug for Korean which is responsible for most of the growth
of the table size.
- As there is no more normalization step, code should now handle both strings
and byte slices as input. Introduced source type to facilitate this.
NOTES
- This change does not handle normalization correctly entirely for contractions.
This causes a few failures with the regtest. table_test.go contains a few
uncommented tests that can be enabled once this is fixed. The easiest is to
fix this once we have the new norm.Iter.
- Removed a test cases in table_test that covers cases that are now guaranteed
to not exist.
Dmitriy Vyukov [Mon, 24 Dec 2012 11:33:32 +0000 (15:33 +0400)]
runtime/race: make test driver print compilation errors
Currently it silently "succeeds" saying that it run 0 tests
if there are compilations errors.
With this change it fails and outputs the compilation error.
Russ Cox [Sat, 22 Dec 2012 22:23:33 +0000 (17:23 -0500)]
cmd/gc: add diagnostic for var, type, const named init
Before this CL, defining the variable worked fine, but then when
the implicit package-level init func was created, that caused a
name collision and a confusing error about the redeclaration.
Also add a test for issue 3705 (func init() needs body).
Russ Cox [Sat, 22 Dec 2012 20:06:28 +0000 (15:06 -0500)]
runtime: ignore failure from madvise
When we release memory to the OS, if the OS doesn't want us
to release it (for example, because the program executed
mlockall(MCL_FUTURE)), madvise will fail. Ignore the failure
instead of crashing.
Rémy Oudompheng [Sat, 22 Dec 2012 18:16:31 +0000 (19:16 +0100)]
cmd/gc: fix wrong interaction between inlining and embedded builtins.
The patch makes the compile user an ordinary package-local
symbol for the name of embedded fields of builtin type.
This is incompatible with the fix delivered for issue 2687
(revision 3c060add43fb) but fixes it in a different way, because
the explicit symbol on the field makes the typechecker able to
find it in lookdot.
remove zerostack compiler experiment; will do at link time instead
««« original CL description
cmd/gc: add GOEXPERIMENT=zerostack to clear stack on function entry
This is expensive but it might be useful in cases where
people are suffering from false positives during garbage
collection and are willing to trade the CPU time for getting
rid of the false positives.
On the other hand it only eliminates false positives caused
by other function calls, not false positives caused by dead
temporaries stored in the current function call.
The 5g/6g/8g changes were pulled out of the history, from
the last time we needed to do this (to work around a goto bug).
The code in go.h, lex.c, pgen.c is new but tiny.
Erik St. Martin [Sat, 22 Dec 2012 16:14:56 +0000 (11:14 -0500)]
regexp: fix index panic in Replace
When using subexpressions ($1) as replacements, when they either don't exist or values weren't found causes a panic.
This patch ensures that the match location isn't -1, to prevent out of bounds errors.
Fixes #3816.
Chris Jones [Sat, 22 Dec 2012 15:00:36 +0000 (10:00 -0500)]
encoding/xml: Marshal ",any" fields
Fixes #3559.
This makes Marshal handle fields marked ",any" instead of ignoring
them. That makes Marshal more symmetrical with Unmarshal, which seems
to have been a design goal.
Note some test cases were changed, because this patch changes
marshalling behavior. I think the previous behavior was buggy, but
there's still a backward-compatibility question to consider.
Dmitriy Vyukov [Fri, 21 Dec 2012 15:12:00 +0000 (19:12 +0400)]
runtime/race: update linux runtime to r170876.
This disables checks for limited address space
and unlimited stack. They are not required for Go.
Fixes #4577.
Rémy Oudompheng [Thu, 20 Dec 2012 22:27:28 +0000 (23:27 +0100)]
cmd/gc: remove an incorrect assertion in escape analysis.
A fatal error used to happen when escassign-ing a multiple
function return to a single node. However, the situation
naturally appears when using "go f(g())" or "defer f(g())",
because g() is escassign-ed to sink.
Joel Sing [Thu, 20 Dec 2012 14:43:19 +0000 (01:43 +1100)]
cgo: enable cgo on openbsd
Enable cgo on OpenBSD.
The OpenBSD ld.so(1) does not currently support PT_TLS sections. Work
around this by fixing up the TCB that has been provided by librthread
and reallocating a TCB with additional space for TLS. Also provide a
wrapper for pthread_create, allowing zeroed TLS to be allocated for
threads created externally to Go.
Joel Sing [Thu, 20 Dec 2012 14:27:50 +0000 (01:27 +1100)]
cmd/[568]l: do not generate PT_TLS on openbsd
The OpenBSD ld.so(1) does not currently support PT_TLS and refuses
to load ELF binaries that contain PT_TLS sections. Do not emit PT_TLS
sections - we will handle this appropriately in runtime/cgo instead.