Log precompile _failures_ as well as errors. Pass log level through to helper process. #457
Conversation
…o helper process.
This allows users to set the debug log level in the process, not just
via the `JULIA_DEBUG=all` environment variable (we _thought_ we had
enabled the debug messages, but we had not! 😬 ).
Also, adds a log statement for precompile statements that _fail_, not
just those that error. For example, this surfaces this failure:
```julia
┌ Debug: Precompilation failed: precompile(Tuple{typeof(Base.deepcopy_internal), Any, Base.IdDict{Any, Any}})
└ @ Main.anonymous none:31
```
Which is a pattern I've noticed in our own precompilation failures,
where Julia seems to be emitting a precompile statement with an abstract
argument (`Any` in this case, but i've seen others) for a method that is
_not_ marked `@nospecialize`, which causes `precompile()` to fail, since
you cannot normally instantiate a method for abstract types.
So that's a separate issue that we need to dig into, but this PR allows
us to identify issues like these. :)
NHDaly
force-pushed
the
nhd-Log-precompile-failures
branch
from
76c29e0 to
6aa2fb2
Compare
|
I would prefer to be able to somehow pass through the Log config if you set I'd love suggestions for that if you have any! :) |
Codecov Report
@@ Coverage Diff @@
## master #457 +/- ##
==========================================
- Coverage 84.91% 84.31% -0.60%
==========================================
Files 2 2
Lines 338 338
==========================================
- Hits 287 285 -2
- Misses 51 53 +2
Continue to review full report at Codecov.
|
| end | ||
| total = length(precompile_statements) | ||
| @info "PackageCompiler: completed \$total precompile statements with \$failures failures, \$errors errors." |
There was a problem hiding this comment.
One problem here is that people think there is something wrong with their code when they see "x errors" but there is really nothing they can do to fix it. So I'm a little bit against adding this as something people see by default.
There was a problem hiding this comment.
Yeah, i had the same concern, and wasn't sure what to do about it.
Maybe a flag, to control reporting precompilation success rate?
Or we could print this to @debug, too, but debug is very noisy and i think we'd like to see this all the time.
There was a problem hiding this comment.
Do you have any suggestions? :) Thanks!
There was a problem hiding this comment.
Maybe we can also print the link to the julia issue that's in the comment, above?
Or do you think I should just move this to a @debug log as well?
NHDaly
force-pushed
the
nhd-Log-precompile-failures
branch
from
5409fb3 to
2e50a35
Compare
This allows users to set the debug log level in the process, not just
via the
JULIA_DEBUG=allenvironment variable (we thought we hadenabled the debug messages, but we had not! 😬 ).
Also, adds a log statement for precompile statements that fail, not
just those that error. For example, this surfaces this failure:
Which is a pattern I've noticed in our own precompilation failures,
where Julia seems to be emitting a precompile statement with an abstract
argument (
Anyin this case, but i've seen others) for a method that isnot marked
@nospecialize, which causesprecompile()to fail, sinceyou cannot normally instantiate a method for abstract types.
So that's a separate issue that we need to dig into, but this PR allows
us to identify issues like these. :)