platform/upstream/ninja.git
9 years agodon't abort()
Scott Graham [Tue, 9 Jun 2015 17:15:11 +0000 (10:15 -0700)]
don't abort()

9 years agopush error to caller and abort on failure
Scott Graham [Tue, 9 Jun 2015 00:17:55 +0000 (17:17 -0700)]
push error to caller and abort on failure

9 years agoFix crash in attempting to canonicalize paths longer than _MAX_PATH
Scott Graham [Mon, 8 Jun 2015 20:21:55 +0000 (13:21 -0700)]
Fix crash in attempting to canonicalize paths longer than _MAX_PATH

9 years agoFix typo in comment.
Nico Weber [Sat, 2 May 2015 21:38:43 +0000 (14:38 -0700)]
Fix typo in comment.

9 years agoMerge pull request #962 from sgraham/pool-use-fix
Nico Weber [Wed, 29 Apr 2015 21:26:54 +0000 (14:26 -0700)]
Merge pull request #962 from sgraham/pool-use-fix

Fix pool use count going unbalanced

9 years agoMerge pull request #743 from nicolasdespres/sigterm
Nico Weber [Mon, 27 Apr 2015 23:42:39 +0000 (16:42 -0700)]
Merge pull request #743 from nicolasdespres/sigterm

Handle SIGTERM properly.

9 years agosimplify & inline
Scott Graham [Mon, 27 Apr 2015 18:12:43 +0000 (11:12 -0700)]
simplify & inline

9 years agoavoid calling ResumeDelayedJobs instead
Scott Graham [Fri, 24 Apr 2015 23:06:55 +0000 (16:06 -0700)]
avoid calling ResumeDelayedJobs instead

9 years agoadd comment
Scott Graham [Fri, 24 Apr 2015 21:25:17 +0000 (14:25 -0700)]
add comment

9 years agoForward interruption signal to child processes.
Nicolas Despres [Wed, 16 Apr 2014 08:08:42 +0000 (10:08 +0200)]
Forward interruption signal to child processes.

9 years agoAllow SIGTERM for interruption.
Nicolas Despres [Wed, 16 Apr 2014 08:05:25 +0000 (10:05 +0200)]
Allow SIGTERM for interruption.

Default signal sent by many other programs (mainly kill(1)) to gently
terminates another one is SIGTERM.

9 years agoFix pool use count going unbalanced
Scott Graham [Thu, 23 Apr 2015 22:54:28 +0000 (15:54 -0700)]
Fix pool use count going unbalanced

9 years agoMerge pull request #958 from rui314/master
Nico Weber [Sun, 19 Apr 2015 03:43:24 +0000 (20:43 -0700)]
Merge pull request #958 from rui314/master

Run more than 34 processes on Win32 if we have 32+ cores.

9 years agoRun more than 34 processes on Win32 if we have 32+ cores.
Rui Ueyama [Fri, 17 Apr 2015 21:14:44 +0000 (14:14 -0700)]
Run more than 34 processes on Win32 if we have 32+ cores.

For compatiblity reason, dwNumberOfProcessors in Win32 is capped at 32.
So even if your machine has more than 32 cores, Ninja spawns at most 34
subprocesses. This patch fixes the issue by using GetNativeSystemInfo,
which returns the system info from Wow64 point of view, instead of
GetSystemInfo.

9 years agoMerge pull request #954 from nico/fix
Nico Weber [Thu, 9 Apr 2015 18:47:28 +0000 (11:47 -0700)]
Merge pull request #954 from nico/fix

Fix an assert (and tests in --debug mode) after #921.

9 years agoFix an assert (and tests in --debug mode) after #921.
Nico Weber [Thu, 9 Apr 2015 18:41:40 +0000 (11:41 -0700)]
Fix an assert (and tests in --debug mode) after #921.

9 years agoMerge pull request #953 from Sushisource/graphivz_fix
Nico Weber [Wed, 8 Apr 2015 19:46:19 +0000 (12:46 -0700)]
Merge pull request #953 from Sushisource/graphivz_fix

Fix backslashes in graphviz causing incorrect rendering on windows.

9 years agoFix backslashes in graphviz causing incorrect rendering on windows.
Spencer Judge [Tue, 7 Apr 2015 21:18:43 +0000 (14:18 -0700)]
Fix backslashes in graphviz causing incorrect rendering on windows.

9 years agoMerge pull request #951 from nico/cyclefix2
Nico Weber [Wed, 1 Apr 2015 21:18:39 +0000 (14:18 -0700)]
Merge pull request #951 from nico/cyclefix2

Don't get stuck on cyclic graphs where one edge has multiple outputs.

9 years agoDon't get stuck on cyclic graphs where one edge has multiple outputs.
Nico Weber [Wed, 1 Apr 2015 15:09:28 +0000 (08:09 -0700)]
Don't get stuck on cyclic graphs where one edge has multiple outputs.

Fixes #934. Plan::AddSubTarget() tracks in want_ if each edge has been
visited and visits every edge only once.  But Plan::CheckDependencyCycle()
worked on nodes however, so if a cycle was entered through an edge with
multiple outputs, ninja would fail to detect that cycle.

Move cycle detection to look for duplicate edges instead of nodes to fix
this.  The extra jump makes CheckDependencyCycle() a bit slower: for a
synthetic build file with 10000 serial build steps, `ninja -n` now takes
0.32s instead of 0.26s before on my laptop.  In practice, projects have
a dependency change length on the order of 50, so there shouldn't be a
noticeable slowdown in practice.  (If this does end up being a problem:
CheckDependencyCycle() currently does O(n) work and is called O(m) times
from AddSubTarget(), so I think cycle checking is O(n^2) in the build
depth.  So instead of worrying about constant overhead, we could use
a set<> instead of a stack<>.  But it doesn't seem to be a problem in
practice.)

9 years agoMerge pull request #950 from nico/cleanup
Nico Weber [Wed, 1 Apr 2015 15:18:49 +0000 (08:18 -0700)]
Merge pull request #950 from nico/cleanup

Simplify Plan's cycle detection code. No behavior change.

9 years agoCleanup: Don't search stack for cycle elements twice.
Nico Weber [Wed, 1 Apr 2015 15:00:27 +0000 (08:00 -0700)]
Cleanup: Don't search stack for cycle elements twice.

The common case is that there is no cycle.  In that case,
CheckDependencyCycle() searched the stack for a dupe from the back,
wouldn't find one, and return false.

If there was a cycle, it would then search again from the front
(probably because the push_back() that used to be here would invalidate
the ri iterator).

Since the push_back() is gone, just search from the front once. No
intended behavior change.

9 years agoCleanup: Make stack a const ref now that it's no longer modified.
Nico Weber [Wed, 1 Apr 2015 14:59:38 +0000 (07:59 -0700)]
Cleanup: Make stack a const ref now that it's no longer modified.

9 years agoCleanup: Don't modify during cycle checking.
Nico Weber [Wed, 1 Apr 2015 14:56:14 +0000 (07:56 -0700)]
Cleanup: Don't modify  during cycle checking.

9 years agoMerge pull request #949 from nico/staterr
Nico Weber [Tue, 31 Mar 2015 19:59:49 +0000 (12:59 -0700)]
Merge pull request #949 from nico/staterr

Let Stat() have an err outparam instead of writing to stderr.

9 years agoLet Stat() have an err outparam instead of writing to stderr.
Nico Weber [Tue, 31 Mar 2015 15:12:12 +0000 (08:12 -0700)]
Let Stat() have an err outparam instead of writing to stderr.

Also check for Stat() failure in a few more places.

This way, ninja doesn't print two "ninja: error: " lines if stat() fails
during a build.  It also makes it easier to keep the stat tests quiet.
Every caller of Stat() needs to explicitly log the error string if
that's desired.

9 years agoMerge pull request #945 from nico/reldocs
Nico Weber [Wed, 25 Mar 2015 21:13:49 +0000 (14:13 -0700)]
Merge pull request #945 from nico/reldocs

make notes on how to update docs more detailed

9 years agomake notes on how to update docs more detailed
Nico Weber [Wed, 25 Mar 2015 19:19:14 +0000 (12:19 -0700)]
make notes on how to update docs more detailed

9 years agoMerge pull request #944 from nico/flags
Nico Weber [Wed, 25 Mar 2015 03:53:06 +0000 (20:53 -0700)]
Merge pull request #944 from nico/flags

Add an opt-in flag to make duplicate edges an error (`-w dupbuild=err`).

9 years agoAdd an opt-in flag to make duplicate edges an error (`-w dupbuild=err`).
Nico Weber [Tue, 24 Mar 2015 21:16:07 +0000 (14:16 -0700)]
Add an opt-in flag to make duplicate edges an error (`-w dupbuild=err`).

This is step 1 on #931.  Duplicated edges will become an error by default in
the future.

9 years agoSwap order of -k and -l in help output, to keep the list alphabetized.
Nico Weber [Tue, 24 Mar 2015 20:37:11 +0000 (13:37 -0700)]
Swap order of -k and -l in help output, to keep the list alphabetized.

9 years agoMerge pull request #943 from nico/ssshh
Nico Weber [Tue, 24 Mar 2015 02:51:35 +0000 (19:51 -0700)]
Merge pull request #943 from nico/ssshh

Make tests quiet again, and prepare for making duplicate build edges an error.

9 years agoMake tests quiet again.
Nico Weber [Tue, 24 Mar 2015 01:18:01 +0000 (18:18 -0700)]
Make tests quiet again.

9 years agoMove warning emission on dupe edges from State to ManifestParser.
Nico Weber [Tue, 24 Mar 2015 01:16:36 +0000 (18:16 -0700)]
Move warning emission on dupe edges from State to ManifestParser.

This will make it easier to optionally make this an error (because
ManifestParser has a way of printing errors), and it'll also make
it easier to make the tests quiet again.

No behavior change.

9 years agoMerge pull request #942 from nico/cyclefix2
Nico Weber [Sat, 21 Mar 2015 15:28:59 +0000 (08:28 -0700)]
Merge pull request #942 from nico/cyclefix2

Don't crash on cyclic references between rule bindings.

9 years agoPreallocate edge node vectors. ~1% faster.
Nico Weber [Sat, 21 Mar 2015 04:21:14 +0000 (21:21 -0700)]
Preallocate edge node vectors. ~1% faster.

9 years agoRecover slowdown for cyclic rule bindings fix.
Nico Weber [Sat, 21 Mar 2015 04:11:31 +0000 (21:11 -0700)]
Recover slowdown for cyclic rule bindings fix.

9 years agoDon't crash on cyclic references between rule bindings.
Nico Weber [Wed, 18 Mar 2015 20:54:16 +0000 (16:54 -0400)]
Don't crash on cyclic references between rule bindings.

Fixes #902.

This dynamically detects cycles.  I like this approach less than
detecting them statically when parsing rules [1], but it has the
advantage that it doesn't break existing ninja manifest files.

It has the disadvantage that it slows down manifest_parser_perftest by
3.9%.

1: https://github.com/martine/ninja/commit/cc6f54d6d436047

9 years agoMerge pull request #940 from nico/moredupes
Nico Weber [Thu, 19 Mar 2015 23:57:54 +0000 (16:57 -0700)]
Merge pull request #940 from nico/moredupes

Another crash fix for duplicate edges. Fixes #939.

9 years agoAnother crash fix for duplicate edges. Fixes #939.
Nico Weber [Thu, 19 Mar 2015 23:42:06 +0000 (16:42 -0700)]
Another crash fix for duplicate edges. Fixes #939.

Patch #933 fixed a crash with duplicate edges by not adding edges to the
graph if all the edge's outputs are already built by other edges.
However, it added the edge to the out_edges of the edge's input nodes
before deleting it, letting inputs refer to dead edges.

To fix, move the check for deleting an edge above the code that adds
inputs.  Expand VerifyGraph() to check that nodes don't refer to edges
that aren't present in the state.

9 years agoMerge pull request #938 from nico/aflclang
Nico Weber [Thu, 19 Mar 2015 18:49:53 +0000 (11:49 -0700)]
Merge pull request #938 from nico/aflclang

Add notes on using afl-fuzz to HACKING.

9 years agoAdd notes on using afl-fuzz to HACKING.
Nico Weber [Tue, 3 Mar 2015 01:11:30 +0000 (17:11 -0800)]
Add notes on using afl-fuzz to HACKING.

9 years agoMerge pull request #937 from nico/statfix
Nico Weber [Thu, 19 Mar 2015 17:13:43 +0000 (10:13 -0700)]
Merge pull request #937 from nico/statfix

Make failing stat() calls abort the build.

9 years agoMake failing stat() calls abort the build.
Nico Weber [Thu, 19 Mar 2015 15:42:46 +0000 (08:42 -0700)]
Make failing stat() calls abort the build.

Fixes #830, fixes #904.

In practice, this either happens with 64-bit inodes and a 32-bit
userspace when building without -D_FILE_OFFSET_BITS=64 in CFLAGS, or
when a filename is longer than the system file length limit.

Since DiskInterface::Stat() returns -1 on error, and Node used -1 on
"stat state unknown", not aborting the build lead to ninja stat()ing the
same file over and over again, until it finally ran out of stack. That's
now fixed.

* Change RecomputeOutputsDirty() to return success instead of dirty
  state (like RecomputeDirty()) and return the dirty state in a bool
  outparam
* Node::Stat()s old return value wasn't used anywhere, change the
  function to return success instead and add an |err| outparam
* Node::StatIfNecessary()'s old return value was used only in one place.
  Change that place to explicitly check status_known() and make
  StatIfNecessary() return success and add an |err| outparam
* Plan::CleanNode() can now fail, make it return bool and add an |err|
  outparam

9 years agoMerge pull request #936 from nico/cleanup
Nico Weber [Thu, 19 Mar 2015 03:56:28 +0000 (20:56 -0700)]
Merge pull request #936 from nico/cleanup

Minor cleanups. No behavior change.

9 years agoAdd a missing &. (No behavior change, only used in tests.)
Nico Weber [Wed, 18 Mar 2015 20:39:26 +0000 (16:39 -0400)]
Add a missing &.  (No behavior change, only used in tests.)

9 years agoEnv should only be about variables. No behavior change.
Nico Weber [Wed, 18 Mar 2015 16:35:57 +0000 (12:35 -0400)]
Env should only be about variables.  No behavior change.

9 years agoMerge pull request #930 from nico/depcolon
Nico Weber [Tue, 17 Mar 2015 01:21:15 +0000 (18:21 -0700)]
Merge pull request #930 from nico/depcolon

On unexpected output in a .d file, rebuild instead erroring.

9 years agoMerge pull request #933 from nico/dupeedgefix
Nico Weber [Sun, 15 Mar 2015 23:49:24 +0000 (16:49 -0700)]
Merge pull request #933 from nico/dupeedgefix

Build self-consistent graphs for dupe edges with multiple outputs.

9 years agoBuild self-consistent graphs for dupe edges with multiple outputs.
Nico Weber [Sun, 15 Mar 2015 23:34:03 +0000 (19:34 -0400)]
Build self-consistent graphs for dupe edges with multiple outputs.

Fixes #867, both the crashes and "[stuck]" issues.

The problem was that a duplicate edge would modify the in_edge of the
outputs of the new build rule, but the edge corresponding to the old
build rule would still think that the in_edge points to itself.
`old_edge->outputs_[0]->in_edge()` would not return `old_edge`, which
confused the scan logic.

As fix, let `State::AddOut()` reject changing in_edge if it's already
set.  This changes behavior in a minor way: Previously, if there were
multiple edges for a single output, the last edge would be kept.  Now,
the first edge is kept.  This only had mostly-well-defined semantics if
all duplicate edges are the same (which is the only case I've seen in
practice), and for that case the behavior doesn't change.

For testing, add a VerifyGraph() function and call that every time any
test graph is parsed.  That's a bit more code than just copying the test
cases from the bug into build_test.cc, but it also yields better test
coverage overall.

9 years agoOn unexpected output in a .d file, rebuild instead erroring.
Nico Weber [Thu, 12 Mar 2015 16:06:50 +0000 (12:06 -0400)]
On unexpected output in a .d file, rebuild instead erroring.

Fixes #417.

9 years agoReject depfiles that don't contain a : after the target name.
Nico Weber [Thu, 12 Mar 2015 15:59:45 +0000 (11:59 -0400)]
Reject depfiles that don't contain a : after the target name.

This is a prerequisite for fixing #417.

9 years agoMerge pull request #928 from nico/novar
Evan Martin [Tue, 10 Mar 2015 23:47:44 +0000 (16:47 -0700)]
Merge pull request #928 from nico/novar

Remove option no longer needed now that we don't use gtest.

9 years agoRemove option no longer needed now that we don't use gtest.
Nico Weber [Tue, 10 Mar 2015 20:45:17 +0000 (13:45 -0700)]
Remove option no longer needed now that we don't use gtest.

9 years agoSimplify. No behavior change.
Nico Weber [Tue, 10 Mar 2015 16:09:49 +0000 (09:09 -0700)]
Simplify.  No behavior change.

9 years agoMerge pull request #917 from pinotree/rlimit
Nico Weber [Mon, 9 Mar 2015 22:50:30 +0000 (15:50 -0700)]
Merge pull request #917 from pinotree/rlimit

subprocess_test: gracefully handle rlim.rlim_cur < kNumProcs

9 years agoMerge pull request #910 from jlnt/master
Nico Weber [Mon, 9 Mar 2015 18:32:25 +0000 (11:32 -0700)]
Merge pull request #910 from jlnt/master

POSIX: detach background subprocesses from terminal.

9 years agoMerge pull request #927 from nico/colorgcc
Nico Weber [Mon, 9 Mar 2015 17:34:08 +0000 (10:34 -0700)]
Merge pull request #927 from nico/colorgcc

Make diagnostics colored with new gccs (4.9+) too.

9 years agoMake diagnostics colored with new gccs (4.9+) too.
Nico Weber [Mon, 9 Mar 2015 17:24:41 +0000 (10:24 -0700)]
Make diagnostics colored with new gccs (4.9+) too.

Both clang and gcc understand -fdiagnostics-color, so use that flag name.
(This will disable colored diagnostics for clangs older than LLVM 3.3,
but that is several years old by now.)

9 years agoMerge pull request #926 from nico/fixlibcpp
Nico Weber [Mon, 9 Mar 2015 04:49:37 +0000 (21:49 -0700)]
Merge pull request #926 from nico/fixlibcpp

Fix build with libc++ after #921.

9 years agoFix build with libc++ after #921.
Nico Weber [Mon, 9 Mar 2015 04:46:34 +0000 (21:46 -0700)]
Fix build with libc++ after #921.

It failed with

    error: field has incomplete type 'EvalString'
    note: in instantiation of exception specification for 'map' requested here
      explicit Rule(const string& name) : name_(name) {}
               ^

9 years agoMerge pull request #921 from mohamed/master
Nico Weber [Mon, 9 Mar 2015 04:44:41 +0000 (21:44 -0700)]
Merge pull request #921 from mohamed/master

Allow scoping rules through subninja

9 years agoMerge pull request #925 from kirbyfan64/master
Nico Weber [Mon, 9 Mar 2015 04:42:41 +0000 (21:42 -0700)]
Merge pull request #925 from kirbyfan64/master

Test for Clang by checking --version

9 years agoTest for Clang by checking --version
Ryan Gonzalez [Mon, 9 Mar 2015 00:09:01 +0000 (19:09 -0500)]
Test for Clang by checking --version

9 years agoMerge pull request #923 from HeisSpiter/master
Nico Weber [Sun, 8 Mar 2015 23:20:37 +0000 (16:20 -0700)]
Merge pull request #923 from HeisSpiter/master

Directly pass the string instead of char * to Truncate util function. It...

9 years agoAdded a new test to illustrate scoped rules
Mohamed Bamakhrama [Sun, 8 Mar 2015 03:13:12 +0000 (04:13 +0100)]
Added a new test to illustrate scoped rules

The new test shows the added value of scoped rules by demonstrating
a multi-level build where a single rules file gets included at all
the levels. By scoping rules, this is possible.

9 years agoDirectly pass the string instead of char * to Truncate util function. It will prevent...
Pierre Schweitzer [Sat, 7 Mar 2015 13:23:41 +0000 (14:23 +0100)]
Directly pass the string instead of char * to Truncate util function. It will prevent useless conversions.

9 years agoAllow scoping rules through subninja
Mohamed Bamakhrama [Sun, 1 Mar 2015 23:57:33 +0000 (00:57 +0100)]
Allow scoping rules through subninja

Ninja didn't support scoping rules through subninja and assumed
a unique rule name in the whole namespace. With this change, this
behavior is changed to allow scoping rules. Two rules can have the
same name if they belong to two different scopes. However, two
rules can NOT have the same name in the same scope.

9 years agosubprocess_test: gracefully handle rlim.rlim_cur < kNumProcs
Pino Toscano [Sat, 28 Feb 2015 10:04:09 +0000 (11:04 +0100)]
subprocess_test: gracefully handle rlim.rlim_cur < kNumProcs

Instead of expecting that the number of open files is well above
kNumProcs, simply "skip" the test in that case, still printing the
message about the test limit (adding the current system limit too).

9 years agoMerge pull request #914 from tzik/typo_fix
Nico Weber [Sat, 7 Feb 2015 07:25:42 +0000 (23:25 -0800)]
Merge pull request #914 from tzik/typo_fix

Typo fix in graph.cc

9 years agoTypo fix in graph.cc
tzik [Fri, 6 Feb 2015 10:45:42 +0000 (19:45 +0900)]
Typo fix in graph.cc

9 years agoMerge pull request #908 from colincross/multipass
Nico Weber [Tue, 3 Feb 2015 21:01:28 +0000 (13:01 -0800)]
Merge pull request #908 from colincross/multipass

Allow manifest rebuild to loop up to 100 times

9 years agoAllow manifest rebuild to loop up to 100 times
Colin Cross [Mon, 26 Jan 2015 21:52:20 +0000 (13:52 -0800)]
Allow manifest rebuild to loop up to 100 times

Ninja generators that bootstrap themselves with Ninja may need
to rebuild build.ninja multiple times.  Replace the 2 cycle loop
with a 100 cycle loop, and print the pass number each time it
restarts.

Original-author: Jamie Gennis <jgennis@gmail.com>

9 years agoPOSIX: detach background subprocesses from terminal.
Julien Tinnes [Thu, 29 Jan 2015 19:33:35 +0000 (11:33 -0800)]
POSIX: detach background subprocesses from terminal.

Put background subprocesses (i.e. subprocesses with no access
to the console) in their own session and detach them from the
terminal.

This fixes martine/ninja#909.

9 years agoMerge pull request #897 from tzik/pending
Nico Weber [Sat, 24 Jan 2015 06:18:34 +0000 (22:18 -0800)]
Merge pull request #897 from tzik/pending

Check pending SIGINT after ppoll/pselect

9 years agoMerge pull request #903 from nico/gcov
Nico Weber [Wed, 21 Jan 2015 01:40:34 +0000 (17:40 -0800)]
Merge pull request #903 from nico/gcov

Document how to run gcov on ninja.

9 years agoDocument how to run gcov on ninja.
Nico Weber [Wed, 21 Jan 2015 01:39:27 +0000 (17:39 -0800)]
Document how to run gcov on ninja.

9 years agoMerge pull request #901 from nico/assertless
Nico Weber [Tue, 20 Jan 2015 05:23:36 +0000 (21:23 -0800)]
Merge pull request #901 from nico/assertless

Remove an incorrect assert.

9 years agoRemove an incorrect assert.
Nico Weber [Tue, 20 Jan 2015 05:20:15 +0000 (21:20 -0800)]
Remove an incorrect assert.

The assert fires on cyclic manifests (found by afl-fuzz).  Since there
was explicit error handing for this case already, just remove the
assert.

9 years agoCheck pending SIGINT after ppoll/pselect
Taiju Tsuiki [Fri, 9 Jan 2015 22:19:11 +0000 (07:19 +0900)]
Check pending SIGINT after ppoll/pselect

ppoll/pselect prioritizes file descriptor events over
a signal delivery. So a flood of events prevents ninja
from reacting keyboard interruption by the user.

This CL adds a check for pending keyboard interruptions after
file descriptor events.

9 years agoMerge pull request #894 from tfarina/has-indent
Nico Weber [Wed, 14 Jan 2015 16:28:58 +0000 (08:28 -0800)]
Merge pull request #894 from tfarina/has-indent

Cleanup: Fix 'hasIdent' variable name/style.

9 years agoCleanup: Fix 'hasIdent' variable name/style.
Thiago Farina [Sat, 10 Jan 2015 22:00:43 +0000 (20:00 -0200)]
Cleanup: Fix 'hasIdent' variable name/style.

Seems more correct to name it has_indent_token and to use the
unix_hacker style.

Signed-off-by: Thiago Farina <tfarina@chromium.org>
9 years agoMerge pull request #889 from nico/del2
Nico Weber [Sun, 4 Jan 2015 18:23:17 +0000 (10:23 -0800)]
Merge pull request #889 from nico/del2

Try to simplify d1e6a29 a bit.

9 years agoTry to simplify d1e6a29 a bit.
Nico Weber [Sun, 4 Jan 2015 02:50:32 +0000 (18:50 -0800)]
Try to simplify d1e6a29 a bit.

9 years agoMerge pull request #888 from nico/del
Nico Weber [Sun, 4 Jan 2015 02:01:20 +0000 (18:01 -0800)]
Merge pull request #888 from nico/del

Remove unused hash<std::string>.

9 years agoRemove unused hash<std::string>.
Nico Weber [Sun, 4 Jan 2015 01:58:15 +0000 (17:58 -0800)]
Remove unused hash<std::string>.

ExternalStringHashMap used to store std::strings long ago.  Since it
doesn't anymore, this specialization isn't needed.  No behavior change.

9 years agoMerge pull request #887 from berenm/master
Nico Weber [Sun, 4 Jan 2015 01:53:57 +0000 (17:53 -0800)]
Merge pull request #887 from berenm/master

Fix compilation errors on Visual Studio 2015 (_MSC_VER 1900)

9 years agoFix compilation errors on Visual Studio 2015 (_MSC_VER 1900).
Beren Minor [Wed, 31 Dec 2014 11:17:21 +0000 (12:17 +0100)]
Fix compilation errors on Visual Studio 2015 (_MSC_VER 1900).

9 years agoMerge pull request #884 from nico/sshh
Nico Weber [Thu, 25 Dec 2014 21:36:16 +0000 (13:36 -0800)]
Merge pull request #884 from nico/sshh

Remove 'Recompacting...' messages.

9 years agoMerge pull request #885 from frasercrmck/zsh_comp_modes
Nico Weber [Mon, 22 Dec 2014 18:42:15 +0000 (10:42 -0800)]
Merge pull request #885 from frasercrmck/zsh_comp_modes

zsh-completion: remove use of 'head' with negative offset

9 years agozsh-completion: remove use of 'head' with negative offset
Fraser Cormack [Sat, 20 Dec 2014 00:12:52 +0000 (00:12 +0000)]
zsh-completion: remove use of 'head' with negative offset

Some systems - like OSX - don't come with a version of head that
supports a negative value for the -n flag. Such systems get a message
such as this when tab-completing ninja's -d flag:

    ninja -dhead: illegal line count -- -1

Using sed instead should be more universally supported.

9 years agoRemove 'Recompacting...' messages.
Nico Weber [Thu, 18 Dec 2014 19:04:14 +0000 (11:04 -0800)]
Remove 'Recompacting...' messages.

Recompacting the build log used to be slow, so it made sense to print this
message.  We then made recompaction much faster, but didn't remove this
message back then.

The deps log only has it because the build log had it.

Since both steps are effectively instant in practice, remove these log messages.

9 years agoMerge pull request #882 from mathstuf/verbose-configure
Nico Weber [Thu, 11 Dec 2014 05:11:32 +0000 (21:11 -0800)]
Merge pull request #882 from mathstuf/verbose-configure

configure: add a verbose mode

9 years agoconfigure: add a verbose mode
Ben Boeckel [Thu, 11 Dec 2014 04:09:38 +0000 (23:09 -0500)]
configure: add a verbose mode

Required for Fedora infrastructure so that the commands used to build
ninja are logged.

9 years agoMerge pull request #881 from nico/depscancycles
Nico Weber [Tue, 9 Dec 2014 20:26:41 +0000 (12:26 -0800)]
Merge pull request #881 from nico/depscancycles

Let DependencyScan::RecomputeDirty() work correclty with cyclic graphs.

9 years agoLet DependencyScan::RecomputeDirty() work correclty with cyclic graphs.
Nico Weber [Mon, 8 Dec 2014 00:02:35 +0000 (16:02 -0800)]
Let DependencyScan::RecomputeDirty() work correclty with cyclic graphs.

RecomputDirty(edge) currently works roughly like this:

      RecomputeDirty(edge):
        LoadDeps(edge)
        for in in edge.inputs:
          if in.StatIfNecessary():
            RecomputeDirty(in.in_edge)  # recurse into inputs
        for out in edge.outputs:
          out.StatIfNecessary()  # mark outputs as visited

It uses the stat state of each node to mark nodes as visited and doesn't
traverse across nodes that have been visited already.  For cyclic graphs
with an edge with multiple outputs on the cycle, nothing prevents an
edge to be visited more than once if the cycle is entered through an
output that isn't on the cycle.  In other words, RecomputeDirty() for
the same edge can be on the call stack more than once.  This is bad for
at least two reasons:

  1. Deps are added multiple times, making the graph confusing to reason
     about.

  2.  LoadDeps() will insert into the inputs_ of an edge that's iterated
      over in a callframe higher up. This can invalidate the iterator,
      which causes a crash when the callframe with the loop over the
      now-invalidated iterator resumes.

To fix this, let RecomputeDirty() mark all outputs of an edge as visited
as the first thing it does.  This way, even if the edge is on a cycle
with several outputs, each output is already marked and no edge will
have its deps loaded more than once.

Fixes the crashes in #875.  (In practice, it turns the crashes into
"stuck [this is a bug]" messages for now, due to the example without
duplicate rules in #867)

9 years agoRename a few iterators. No behavior change.
Nico Weber [Sun, 7 Dec 2014 22:49:49 +0000 (14:49 -0800)]
Rename a few iterators. No behavior change.

It confused me that the iterator iterating over `outputs_` was called
`i` -- this always made me think of "input", not "iterator".

Call iterators over edge outputs "o", iterators over edge inputs "i",
iterators over node input edges "oe", and general iterators over edges
"e".

9 years agoremove two unneeded `explicit`s
Nico Weber [Sun, 7 Dec 2014 22:42:12 +0000 (14:42 -0800)]
remove two unneeded `explicit`s

9 years agoMerge pull request #703 from ovinn/master
Nico Weber [Thu, 4 Dec 2014 18:05:44 +0000 (10:05 -0800)]
Merge pull request #703 from ovinn/master

Added *.obj ignore rule

9 years agoMerge pull request #835 from rainlabs-eu/fix-cygwin-compatibility
Nico Weber [Thu, 4 Dec 2014 18:03:49 +0000 (10:03 -0800)]
Merge pull request #835 from rainlabs-eu/fix-cygwin-compatibility

Fixed cygwin compatibility (issue #806)