Shuffle the responsibility for creating the TaskQueue out of the
Compilation's internal job state object and into the driver. Expose
a builder convenience function that handles the argument parsing.
When generating a compiler invocation in driver::createCompilerInvocation()
we end up using filelists if the number of inputs is > 128 (to work around
command line arg limits). We never actually write them out though, and so
fail when parsing the frontend arguments that reference them.
As this function is called frequently by SourceKit and command line limits
aren't a concern here, this patch makes the 128 threshold value configurable
via a new -driver-filelist-threshold option. This is set to its maximum value
in driver::createCompilerInvocation() to ensure filelists aren't used. This
new option makes the existing -driver-use-filelists (that forces filelists to
be used) redundant as it's now equivalent to -driver-filelist-threshold=0.
Resolves rdar://problem/38231888
* In full compilation '-c' with '-emit-module', output duplicated build
record file for full compilation *and* emit module ('~moduleonly') mode.
* In '-emit-module' only mode, use '~moduleonly' build record.
Previously, Actions were responsible for freeing their inputs...
except for the ones that weren't. Or the ones that were supposed
to, but then they needed to share an input, so they couldn't anymore.
If this sounds ridiculous, you're right; now Actions are just
immediately allocated and owned by the Compilation.
The graph structure of the actions is still useful for some things; in
particular, "top-level" actions get to put their outputs somewhere
permanent rather than TMPDIR. But I expect these things to get cleaned
up in the future too.
This should make it easier to rerun crashed jobs that use filelists;
previously you'd have to run the top-level driver command again with
-save-temps. I didn't want to save /all/ temporary files because that
often includes things like .o files, which could fill up your disk
pretty quickly. But we can always tweak this later.
Generate frontend commands with -filelist in them. This isn't actually
implemented yet, but we can start testing at this point.
Part 1 of https://bugs.swift.org/browse/SR-280.
Previously jobs had to grovel this information out of the raw argument
list, which dropped the types we had inferred on input files. This
makes things more consistent across the compiler, though arguably we
should be able to designate "primary" and "non-primary" inputs on a
per-action basis rather than resorting to "global" state.
Use this new information to stop passing object file inputs to the
Swift frontend, fixing rdar://problem/23213785.
The list wouldn't have to live on the Compilation, but I'm going to use
it to fix SR-280 / rdar://problem/23878192 as well.
DerivedArgList has a pointer to the InputArgList it came from, so we can't
just std::move it around. Put most of the driver back the way it was, with
small changes to clarify ownership.
Swift SVN r31811
This was just a wrapper around SmallVector that optionally owned the Job pointers
in it. Now that all Jobs are owned by the Compilation, we don't have to worry
about this any more.
No functionality change.
Swift SVN r29668
Previously, we would process all of a job's dependencies separately before
even scheduling it, and we wouldn't interleave dependencies from different
jobs. This meant (a) more overhead than necessary, and more importantly
(b) -embed-bitcode builds weren't being parallelized.
rdar://problem/21129029
Swift SVN r29665
...not if it's newer than its output .o file. This handles cases where the
object file is generated too quickly (rdar://problem/19404140) or when you
revert to a previous version of the file, mtime intact (rdar://problem/19720146).
There's a lot of test churn here; the only real new test is the backwards
mtime update in one-way.swift.
Swift SVN r29584
If certain command-line arguments change, the results of the last
compilation aren't reusable, i.e. we can't do an incremental build.
Do a full rebuild when we detect that this happens.
(Which command-line options? Conservatively assume all of them, /except/
those with the new DoesNotAffectIncrementalBuild flag in Options.td.)
Swift SVN r24385
This is important because we might get part-way through the full
compilation, overwriting swiftdeps files as we go, and then encounter an
error. We don't want to lose information about any decls that have been
removed since the previous compile, so we propagate forward the information
we already have by saving it to a "build record" file.
More simply, this is necessary to track when a file is removed from a target.
The next commit will handle reading in this file at the start of a build.
Swift SVN r23968
