Currently `UnifiedStatsReporter::flushTracesAndProfiles`
can kick off requests when computing the source
ranges for collected entities, which will try to
record additional stats about the requests.
This currently happens to work without issue,
but #29289 bumped the counters down slightly such
that the vector storing the stats now performs a
re-allocation when we do a reentrant stat entry.
This then caused a use-after-free as we try to
continue iterating over the old buffer.
Fix this issue by refusing to record any new stats
while we're flushing out the ones we've already
recorded.
Windows requires a handle to get memory usage, so do a slight refactor
to collect the child's memory usage as it exits instead of as the parent
is cleaning up.
Adds a stat to SILInstruction's transferNodesFromList to record the
number of times an instruction is transfered to another block. This is
the only way I can think of to detect quadratic behavior of passes
that split basic blocks.
This counts the number of leaf scopes we reach while solving the
constraint sytem, and is a much better measure of the growth of
unnecessary work than the total number of scopes opened.
There were two tests where I had a difficult time getting scale-test
to fit the curve even after adjusting some of the parameters, so I've
left those to use the old stat for now.
When linking on Linux, we would fail with unresolved symbol references
to swift::FrontendStatsTracer::getTraceFormatter<T>. The use of the
types, which are defined in swiftSIL occur in swiftBasic. Provide
inline definitions of the constructors which cause the dependency on
some environments (e.g. Linux).
This helps disambiguate files that might otherwise be hard to sort through
if multiple runs output stats together in a single directory. The names
don't have to be perfect, just contain sufficient hints (and be parseable)
to differentiate module, arch, opt and output-type variation in jobs.