Precompile for speed

[staticfloat]

Also increase relocatability

[KristofferC]

I have never seen precompile give any speed benefit except if used in combination with sysimg building. Do you have any numbers here or is the purpose actually to make BP faster when used in sysimgs?

[KristofferC]

Also not that you are including all anonymous functions here, so as soon as you make a change in a closure or something, this will likely break since the symbol will be gone.

[staticfloat]

Yeah, I figured that this would break as soon as someone breathes on the codebase. But that doesn’t particularly annoy me; it’s easy to regenerate the file. I measure about 50ms of speedup (So that’s a 10% overall speedup). It looks like the `__init__()` function spends that time compiling itself and e.g. `probe_platform_engines()`. With this change, there’s essentially no compilation done at all during startup, it’s all just 400ms spent reading serialized code in from disk. -E On October 6, 2018 at 10:48:07 PM, Kristoffer Carlsson ( notifications@github.com) wrote: Also not that you are including all anonymous functions here, so as soon as you make a change in a closure or something, this will likely break since the symbol will be gone. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#126 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAH_aKsUKvBI1Kk3RlvBQ70OPJv_V_3Jks5uiMKngaJpZM4XLSdl> .

[staticfloat]

Yeah, I figured that this would break as soon as someone breathes on the codebase.  But that doesn’t particularly annoy me; it’s easy to regenerate the file. I measure about 50ms of speedup (So that’s a 10% overall speedup).  It looks like the `__init__()` function spends that time compiling itself and e.g. `probe_platform_engines()`.  With this change, there’s essentially no compilation done at all during startup, it’s all just 400ms spent reading serialized code in from disk. -E On October 6, 2018 at 10:48:07 PM, Kristoffer Carlsson (notifications@github.com) wrote: Also not that you are including all anonymous functions here, so as soon as you make a change in a closure or something, this will likely break since the symbol will be gone. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.

[codecov-io]

Codecov Report

Merging #126 into master will decrease coverage by 0.17%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master     #126      +/-   ##
==========================================
- Coverage    88.4%   88.23%   -0.18%     
==========================================
  Files           8        8              
  Lines         647      646       -1     
==========================================
- Hits          572      570       -2     
- Misses         75       76       +1

Impacted Files	Coverage Δ
src/BinaryProvider.jl	100% <100%> (ø)	⬆️
src/Prefix.jl	90.6% <0%> (-0.62%)	⬇️
src/PlatformEngines.jl	71.95% <0%> (-0.17%)	⬇️
src/OutputCollector.jl	93.61% <0%> (-0.07%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update cb403e6...302ffa8. Read the comment docs.

[staticfloat]

Welp, that didn't take long to get annoying. :P I removed everything that had a name starting with #.

[KristofferC]

You could always just generate it for a release.

[staticfloat]

Yes, well another annoyance is that some anonymous functions don't exist on some platforms, so the precompile.jl file has to be the intersection of snoop runs on all different platforms. :/

It's okay, with these few lines I already get most of the benefit, so no biggie.