Precompile errors should attach the underlying exception

[BioTurboNick]

Currently, failed precompile just uses error when it encounters an exception. Probably better for it to attach the underlying exception that caused it to fail?

julia/base/loading.jl

Lines 1430 to 1467 in 66c3375

	try
	close(tmpio)
	p = create_expr_cache(pkg, path, tmppath, concrete_deps, internal_stderr, internal_stdout)
	if success(p)
	# append checksum to the end of the .ji file:
	open(tmppath, "a+") do f
	write(f, _crc32c(seekstart(f)))
	end
	# inherit permission from the source file
	chmod(tmppath, filemode(path) & 0o777)
	# Read preferences hash back from .ji file (we can't precompute because
	# we don't actually know what the list of compile-time preferences are without compiling)
	prefs_hash = preferences_hash(tmppath)
	cachefile = compilecache_path(pkg, prefs_hash)
	# prune the directory with cache files
	if pkg.uuid !== nothing
	entrypath, entryfile = cache_file_entry(pkg)
	cachefiles = filter!(x -> startswith(x, entryfile * "_"), readdir(cachepath))
	if length(cachefiles) >= MAX_NUM_PRECOMPILE_FILES[]
	idx = findmin(mtime.(joinpath.(cachepath, cachefiles)))[2]
	rm(joinpath(cachepath, cachefiles[idx]))
	end
	end
	# this is atomic according to POSIX:
	rename(tmppath, cachefile; force=true)
	return cachefile
	end
	finally
	rm(tmppath, force=true)
	end
	if p.exitcode == 125
	return PrecompilableError()
	else
	error("Failed to precompile $pkg to $tmppath.")
	end

[DilumAluthge]

I believe you can just get the exception directly by doing e.g. import Foo, right?

[BioTurboNick]

Possibly, though I believe it's proper exception handling to propagate originating errors. Once #40642 is merged, which allows the previous error to be re-displayed, this will become more apparent as only the stack trace from compilecache would be accessible, and not the trace from the actual problem.

[KristofferC]

I believe you can just get the exception directly by doing e.g. import Foo, right?

I think this issue is in fact about normal precompilation that occurs during import Foo and not Pkg parallel precompilation.

[vtjnash]

There is no distinct error there though, since it is just a failed precompile process, and that is all it knows. Also that outer packages should not be trying to delve into recursively generated errors, and just treat is as: using failed.