Failure on macOS in master

[certik]

There is an error in master on macOS 9:

https://github.com/fortran-lang/stdlib/runs/1532433788?check_suite_focus=true

I first noticed it in #263, but it couldn't have been caused by that PR. The first time it appeared seems to be after merging #260, but probably also unrelated.

We need to fix it.

[jvdp1]

It is related to the PR #239 .But the issue appears much later. I (and I believe @wclodius2 too) already try to fix it. I never could reproduce it on my computer (Fedora) with both gfortran and ifort 19.
Re-running the actions usually usually result in all passed actions. It seems to be quite random, and is only in the in-tree built apparently.

[certik]

It needs to be debugged. I think it fails somewhere around here:

stdlib/src/tests/bitsets/test_stdlib_bitset_large.f90

Line 265 in 4bf52a4

call set3 % read_bitset(unit)

[certik]

It actually fails for me on my laptop. Bisecting the error.

[jvdp1]

Note that for the PR #261 (unrelated to bitsets), all tests passed.

It actually fails for me on my laptop. Bisecting the error.

Good! Hopefully you can fix it.

[Romendakil]

Things that come to my mind are race conditions around the test.txt file, or a problem somewhere maybe with the case-insensitiveness of the default macOS file system.

[certik]

Here is the backtrace:

 Test bitset I/O: input, read_bitset, output, and write_bitset
STDLIB_BITSETS % READ_BITSET: Invalid character in bitset literal in UNIT
ERROR STOP A character string had an invalid character.

Error termination. Backtrace:
#0  0x7f31ac1032ed in ???
#1  0x7f31ac103ed5 in ???
#2  0x7f31ac104f86 in ???
#3  0x561872a8c764 in __stdlib_bitsets_MOD_error_handler
	at /home/ondrej/repos/stdlib/src/stdlib_bitsets.f90:2161
#4  0x561872a93818 in __stdlib_bitsets_MOD_read_bitset_unit_large
	at /home/ondrej/repos/stdlib/src/stdlib_bitsets_large.f90:1178
#5  0x561872a897b4 in test_io
	at /home/ondrej/repos/stdlib/src/tests/bitsets/test_stdlib_bitset_large.f90:37
#6  0x561872a8b5ab in test_stdlib_bitset_large
	at /home/ondrej/repos/stdlib/src/tests/bitsets/test_stdlib_bitset_large.f90:14
#7  0x561872a8b5ed in main
	at /home/ondrej/repos/stdlib/src/tests/bitsets/test_stdlib_bitset_large.f90:2

I use gfortran and Ubuntu 18.04.

[jvdp1]

I use gfortran and Ubuntu 18.04.

Strange. I just tried again on my computer (Fedore 31) with gfortran, and I couldn't reproduce it. Good luck! The backtrace doesn't seem to be useful.

[certik]

I made a mistake. I removed the call test_string_operations() in attempts to simplify it, which actually causes an error, but a different error (the sets were not initialized).

[certik]

I am debugging it at #265.

[certik]

Fixed by #267. The master (7b24642) now passes. But let's keep watching it in case the problem is still there.

[wclodius2]

Now that Ondrej has apparently fixed it it doesn’t matter, but I use a MacBook Pro with gfortran 8.1.0 and 10.2.0 and ifort 18.0.3 installed and was never able to reproduce the error with any of the three compilers. From the naming I suspect the Mac that was failing was running gfortran 9.something.

…

On Dec 10, 2020, at 11:58 AM, Jeremie Vandenplas ***@***.***> wrote: It is related to the PR #239 <#239> .But the issue appears much later. I (and I believe @wclodius2 <https://github.com/wclodius2> too) already try to fix it. I never could reproduce it on my computer (Fedora) with both gfortran and ifort 19. Re-running the actions usually usually result in all passed actions. It seems to be quite random, and is only in the in-tree built apparently. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#264 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/APTQDOWT3REFZSAXXAKIXYLSUEK47ANCNFSM4UVNFIMQ>.