Fix empty sysimage test on 1.5.3. Fix app test on nightly #476
Conversation
Codecov Report
@@ Coverage Diff @@
## master #476 +/- ##
==========================================
+ Coverage 77.98% 82.69% +4.70%
==========================================
Files 2 2
Lines 368 393 +25
==========================================
+ Hits 287 325 +38
+ Misses 81 68 -13
Continue to review full report at Codecov.
|
|
On 1.5.3, the final empty sysimage test fails (which doesn't fail on 1.5.2 or nightly) julia> tmp = mktempdir()
julia> sysimage_path = joinpath(tmp, "empty." * Libdl.dlext)
julia> foreach(x -> touch(joinpath(tmp, x)), ["Project.toml", "Manifest.toml"])
julia> create_sysimage(; sysimage_path=sysimage_path, incremental=false, filter_stdlibs=true, project=tmp)not filtering stdlibs succeeds julia> create_sysimage(; sysimage_path=sysimage_path, incremental=false, filter_stdlibs=false, project=tmp) |
IanButterworth
force-pushed
the
ib/artifact_app
branch
from
420600d to
fec20d4
Compare
IanButterworth
changed the title
Not much help, but looking at the two segfaults reported in Windows nightly CI at https://github.com/JuliaLang/PackageCompiler.jl/pull/476/checks?check_run_id=1544435610, they both crash in Could this be related to changes made at JuliaLang/julia#36588 or followup JuliaLang/julia#38160? That is new on 1.6, deals with loading, and has some Windows specific code. Perhaps @staticfloat can tell. Some linking updates were already done in #471. |
|
It might be related. It depends on what library it's trying to load. Is there a way to either instrument the build or catch it in |
| @@ -1,70 +0,0 @@ | |||
| # This file is machine-generated - editing it directly is not advised | |||
There was a problem hiding this comment.
Was this removed because it wasn't compatible between julia versions?
|
I'm not sure how to find it out. It is using a custom system image created before that in the test, that holds Example.jl. Couldn't get it to work on a debug build. Did get this, but it doesn't give much extra information: |
|
We just merged a PR that changes some areas of this on Windows; can you try again and see if the latest nightly still has issues here? |
|
I still see the same failure on the latest nightly: |
|
@visr can you try with JuliaLang/julia#39009 |
|
Merging this as a improvement. |
|
@staticfloat unfortunately that PR doesn't resolve this issue: |
|
Can you open a new issue with steps on how to reproduce that? Thanks! |
Tidied up summary:
On 1.5.3
temp_cleanup_purgewas failing to compile at exit during generation of a blank sysimage, so this clears the_atexitactivites during the sysimage generation subprocess. This is more of a bandaid than a fix. I couldn't figure out why it was failingThe checked-in Manifest for the test app was causing stdlib jll problems, so the Manifest has been deleted and an instantiate step has been added to
create_app. On nightly ubuntu a deadlock was occurring when autoprecomp was enabled after instantiate, but not on macos. I couldn't figure out why, but I disabled autoprecompA windows failure remains during the first app test. I have no clues on that one (apart from it appearing to be another
_atexit-related issue)