fix: exla rpath

[polvalente]

No description provided.

[polvalente]

small fix needed due to the addition of the version to the EXLA lib compile path

[polvalente]

There's some weird interplay between how EXLA looks up the rpath if it's the main app or if it's a library.
Just copying over the unversioned cache .so seems to be the simplest solution.

This probably doesn't solve the problem of dealing with a stale libexla.so, but the new env var, even in partial mode, will deal with rebuilding everything for a new version

[jonatanklosko]

I would do this:

case System.get_env("EXLA_FORCE_REBUILD") do
  value when value in [nil, "0", "false"] ->
    :none

  "partial" ->
    :partial

  value when value in ["true", "1"] ->
    :full

  other ->
    ...
end

A cond also works.

[polvalente]

I think the readability is either equal or worse in that alternative, given it's a small set of possible values

[jonatanklosko]

With cond readability should be just as good as today and it should be easier to scan through. Your call :)

[josevalim]

Agreed this would suit cond perfectly.

[josevalim]

There is an issue here in that the .so can be huge, back then it was 300MB, so duplicating that in disk can be too much.

[josevalim]

Maybe we keep the .o versioned and not the final one. This should be enough for make to trigger a rebuild whenever the version changes.

[polvalente]

There is an issue here in that the .so can be huge, back then it was 300MB, so duplicating that in disk can be too much.
I checked and it was like 2MB or something.

I'll keep libexla.so unversioned for simplicity then.
The main problem was stale intermediate .o, and that's already solved