fix #16206, nim r / nim -r recompiles if cwd changes

[timotheecour]

fix #16271

nim r / nim -r recompiles if cwd changes

I'm hitting same bug as original reporter of #16271, eg with nim r testament/testament args...; this PR fixes that in the simplest possible way, that is robust even if pessimistic (recompile on cwd change). Future PR's can add more logic to be robust and less pessimistic, see comment in PR.

future work

• -d:nimBetterRunIgnoreCwd (see pr_fix_nimcache_nimrun_16271_followup_nimBetterRunIgnoreCwd wip PR)
• robust approach described inside PR comment (it has other advantages which is caching more than 1 compilation, using a LRU cache to avoid blowing up disk usage)

[Clyybber]

This means that doing

nim c file
cd ..
nim c folder/file

triggers a recompilation. This is too eager IMO.
Instead those flags that implicitly depend on the CWD should be checked to make sure the CWD is added to the respective path before checking for changes.

[timotheecour]

Instead those flags that implicitly depend on the CWD should be checked to make sure the CWD is added to the respective path before checking for changes.

I don't know of a reliable way to do that, so many flags can depend on CWD:

# examples:
nim r --lib:lib main
--path:.
--outdir:dir
-o:bin/main

# also:
--import
--include
--cincludes
--nimcache
--clibdir
--clib
--docRoot
--NimblePath
--excludePath

Furthermore, this would involve among other things understanding and parsing clang flags and linker flags, eg:

nim r --passl:-Wl,-rpath,lib --hotcodereloading:on /pathto/main.nim

or nim's own flags: --docCmd:"--lib:lib"

I also don't know which other flags (or error messages, depending on provided flags) implicitly depend on cwd, jsondoc ctags buildIndex genDepend, genDeps...

The robust approach I have in mind is described inside PR comment, and can be done later (it has other advantages which is caching more than 1 compilation, using a LRU cache to avoid blowing up disk usage).

That said, I'd be totally ok with an opt-in flag though, eg -d:nimBetterRunIgnoreCwd (named after existing -d:nimBetterRun, or a better name) which would use a best effort approach (starting with expanding input project file).

[timotheecour]

@Clyybber in light of #16354 i think we should merge this first, and once #16354 is fixed, introduce an opt-in -d:nimBetterRunIgnoreCwd which would use a best effort (optimistic) algorithm to decide whether to rebuild.

[ringabout]

Related: #16531