search

gfngfn / SATySFi

A statically-typed, functional typesetting system
GNU Lesser General Public License v3.0
1.16k stars 83 forks source link

Travis started to fail #46

Closed qnighy closed 6 years ago

qnighy commented 6 years ago

Currently #44 (build #929) and #45 (build #930) fail (for seemingly the same reasons). As #44 is a trivial change, they must be spurious errors.

I couldn't figure out the exact cause. Seems to be related to core_kernel.

mt-caret commented 6 years ago

I get the same error message as https://travis-ci.org/gfngfn/SATySFi/builds/348081919 while packaging SATySFi for Nix; I would love to know how to fix it.

Logs: https://gist.github.com/mt-caret/26689bcc94a81274081ff3927fa1d9c5

I'm at 042a8aaf3b61785b8edac7d799067ad758ed2fd7 and I'm getting this error since 26aa766d49ddc297741d72ddb2ba8d40457cc61c so the error might be originating somewhere else...

qnighy commented 6 years ago

@mt-caret #47 might help you.

qnighy commented 6 years ago

I tried to bisect dune (formerly jbuilder) between 1.0+beta17 and 1.0+beta18 with the following test script:

opam reinstall -y jbuilder
opam reinstall -y core_kernel

Bisect reported ocaml/dune@9e4cc843704262b3cb563096ae7df128d0ea2bff as the first bad commit.

mt-caret commented 6 years ago

I can confirm, https://github.com/gfngfn/SATySFi/pull/47 indeed fixes the error. Thank you!

An upstream bug, then?

qnighy commented 6 years ago

Probably. I'm investigating further.

qnighy commented 6 years ago

The bisect procedure above was wrong; now I pinned jbuilder and satysfi to local repos and bisected on jbuilder with the following script:

opam reinstall -y jbuilder
opam reinstall -y core_kernel
opam reinstall -y satysfi

Bisect reported ocaml/dune@b3838284c6f28aaa0afabf865ce126c20e2f3616 as the first bad commit.

mt-caret commented 6 years ago

There's a PR that might be addressing the issue: https://github.com/ocaml/dune/issues/567

qnighy commented 6 years ago

Confirmed locally that the current master (ocaml/dune@e4aac2da97c069a914c9fb6c06697b5a99145d17) does not cause the Core_kernel__ problem. I think it is good to pin to 1.0+beta17 until 1.0+beta19 is released.