This shifts explorer's file_test to use ExplorerMain in order to be able
to pass arguments similar to how the command line would. It also means
that various output wrapping done by the command line is shared, no
longer copied by file_test.
In order to help make this work, I plugged vfs::FileSystem into
explorer, similar to how we're doing this on the toolchain side (might
try to share more code later). I was having trouble getting an overlay
working, I think due to how paths are specified -- but due to the issues
in fixing this, I'm just loading the prelude into the InMemoryFilesystem
for now.
Under this approach, I'm unifying more of the file versus string
handling. I think we should shift towards an approach where, like
toolchain code, anyone trying to use Explorer should use a vfs
implementation to supply code. This should reduce the number of distinct
code paths which we're maintaining/testing.
Short-term, this allows the trace testdata to be merged into file_test
with less special casing, using ARGS:. Long-term, it means that the
file_test knows the ARGS to pass to generate checks to match against,
which is the direction I'm planning for autoupdate.
This is intended to address currently flaky timeouts that are likely caused by the size of the prelude. I'm addressing a performance bottleneck in AnalyzeProgram with trace output. Trying to omit prelude traces reduces most trace output significantly, and I think it'll scale better as the prelude size increases.
The basic mechanics here are:
- In order to consistently track whether tracing is on, I've added a TraceStream class, explorer/interpreter/trace_stream.h.
- The AST now has a num_prelude_declarations field, so that it's provided where the boundary is.
- In order to mark where we try to skip prelude output, I've added calls to set_in_prelude in type_checker.
- In exec_program, I just use num_prelude_declarations directly to skip over.
- Everywhere checks TraceStream::is_enabled before printing, similar to the std::optional check that was previously used.
This does add some timing output in order to better diagnose where slowness is coming from, when tracing. It also adds "verbose" targets to make it easier to get the trace output.
So for example, here's a timing for zero.carbon:
```
Timings:
- Parse: 13ms
- AddPrelude: 25ms
- AnalyzeProgram: 116ms
- ExecProgram: 12ms
```
If I make a small change to just not set skipping_prelude (essentially getting back to current output):
```
- Parse: 13ms
- AddPrelude: 25ms
- AnalyzeProgram: 2359ms
- ExecProgram: 57ms
```
Thus in this trivial example, I'm eliminating about 95% of the execution time.
Note this approach could still be refined in a few ways:
- We could add a flag to allow overriding in_prelude. It should be a small amount of work after this change. But it's a little consistent with how parser_debug works, that it won't print prelude output by default (unless there's an error).
- Execution could skip messages involving initialization of globals declared in the prelude. This is a little noisy right now, but I don't think it's significant for performance because ExecProgram is tiny.
- Once files are more separated, we should be able to change the num_prelude_declarations/set_in_prelude approach.
Co-authored-by: Richard Smith <richard@metafoo.co.uk>
Jon Ross-Perkins
