Segfault when running the testsuite with `AGDA_BIN` that is not an absolute path

[andreasabel]

Segfault when running the testsuite with `AGDA_BIN` that is not an absolute path

[andreasabel]

It seems the segfault appears if AGDA_BIN is not an absolute path:

$ dist-2.6.5-ghc-9.4.7/build/agda-tests/agda-tests -i -j1 --regex-include all/Fail
... AGDA_BIN not set
... (segfault)

$ make AGDA_BIN=$(which agda-quicker) fail
(works)

$ AGDA_BIN=$(which agda-quicker) dist-2.6.5-ghc-9.6.3/build/agda-tests/agda-tests -i -j1 --regex-include all/Fail
(works)

The last change around AGDA_BIN was

• #6850

@lawcho: Do you see whether your changes could have triggered this bug?

[andreasabel]

Smells like this bug (fixed in process-1.6.18.0 of September 2023):

• haskell/process#295

[andreasabel]

I tested with GHC 9.8 which ships process-1.6.18.0. Indeed the bug is gone there, while the same experiments with GHC 9.6.3 and 9.4.7 (both ship 1.6.17.0) reproduce the segfault.
So we could sit this out or work around this along the lines of haskell/process#296, by supplying a cwd that is not Nothing.

[andreasabel]

Here is the results for cabal run agda-tests -O0 -w ${GHC} for the following values of GHC, using the Agda 2.6.4 codebase (tag 2.6.4-release-base).

GHC	process	segfault
9.0.2	1.6.13.2	no
9.2.8	1.6.16.0	yes
9.4.7	1.6.17.0	yes
9.6.3	1.6.17.0	yes
9.8.0	1.6.18.0	no

[andreasabel]

Turns our that Agda 2.6.3 didn't have this problem with the testsuite, so it is a regression in 2.6.4, albeit a harmless (devx only) one.