Invalid path to GAP executable in sysinfo

[fingolfin]

We currently do this in regenerate_gaproot():

sysinfo["GAP"] = joinpath(gaproot_mutable, "bin", "gap.sh")

However, there is no bin/gap.sh in gaproot_mutable.

This is irrelevant most of the time, but when building packages it may be important, as some package use that variable in their build system to find a GAP executable, e.g. so they can build their manuals.

So we could use create_gap_sh() or a variant of that to provide that. (A variation because at the very least, we'd want to silence the @info messages there. But also this command creates a fresh Julia environment which is comparatively slow. Perhaps instead we could just generate a script that uses the currently active environment, or something like that?)

[fingolfin]

One place where this came up is in this comment: #1079 (comment)

Also in PR #1067 (see comments in the diff of that PR) we see that some GAP packages (mostly from the homalg project?) would like to launch GAP processes and fail if they can't find one. However, those don't use sysinfo.gap, so it is a somewhat different but still highly related issue.

[lgoettgens]

So we could use create_gap_sh() or a variant of that to provide that. (A variation because at the very least, we'd want to silence the @info messages there. But also this command creates a fresh Julia environment which is comparatively slow. Perhaps instead we could just generate a script that uses the currently active environment, or something like that?)

This approach (using the currently active environment) seems very prone to race conditions to me. If multiple julia sessions with GAP.jl loaded share the same gaproot_mutable (as it is currently the case, cf. #1079), to which "currently active environment" should the file in joinpath(gaproot_mutable, "bin", "gap.sh") point to?