rescript-compiler
rescript-compiler copied to clipboard
rescript-lang
Likely race conditions when core library files are modified
See for example https://github.com/rescript-lang/rescript-compiler/pull/5539#issuecomment-1180325550
Commands behaving non-deterministically, i.e. run it again and you might get a different result:
1 ./scripts/ninja.js clean, config , build
2 node scripts/install -force-lib-rebuild
It could be that running with ninja -j 1 makes this more deterministic. If so that would confirm.
On the linked pr, command 2 seems to be wrong approximately half of the times
I have run node scripts/install -force-lib-rebuild 10 times on https://github.com/rescript-lang/rescript-compiler/pull/5539 with -j 1 and not once has it given a wrong result.
Another observation is that clean / config / build would sometime produce garbled js files with the closing } appearing twice.
Also every single time: doing build a second time runs about 700 steps again. The third time it does nothing.
The problem seems to be that there are two copies of the Js module, and you only modified one of them in #5539.
The duplicate Js module was introduced here: https://github.com/rescript-lang/rescript-compiler/commit/148d57f752c3cfbc30e1dba57c0e2d1dd7c0d23f
The duplicate
Jsmodule was introduced here: 148d57f
Looks like it might have been only introduced to hide certain types used by the compiler and not exposed to the user.
Still, there are race conditions. This file duplication is simply something that makes the race conditions obvious.
Turns out my problem wasn't race conditions, rather actual compile errors caused by changes on the master branch. Sorry for jumping to conclusions.
