load_library() panics on errors when err=0

[apaz-cli]

load_library with err=0 now panics on errors, provided that the file is openable. It used to never panic on errors, leading to confusion between when cases the libjuliacodegen library had been intentionally removed and when it tried but failed to load.

Fixes #47027

[apaz-cli]

Amended for posix, windows was also wrong. I misread the docs initially.

[apaz-cli]

Removed extra whitespace that I accidentally added. Added a test, and CI is passing.

[giordano]

$ julia-1.8.2/bin/julia -q

SYSTEM: caught exception of type ^C^C^C^C^C^C^CWARNING: Force throwing a SIGINT
fatal: error thrown and no exception handler available.
InterruptException()
$ julia-57f9db2279/bin/julia -q
ERROR: Unable to load dependent library /home/dc-gior1/tmp/julia-57f9db2279/bin/../lib/julia/libjulia-codegen.so.1
Message:/usr/local/software/master/gcc/8/lib64/libstdc++.so.6: version `GLIBCXX_3.4.26' not found (required by /home/dc-gior1/tmp/julia-57f9db2279/bin/../lib/julia/libjulia-codegen.so.1)

Much better error, yay!

[gbaraldi]

@giordano maybe this can show what's going on with windows on gcc12?

[giordano]

I was thinking (and hoping) the same!

[vtjnash]

That was the goal

[DilumAluthge]

This PR broke the x86_64-w64-mingw32 tests.

[staticfloat]

It looks to me like we might be missing a UTF16 -> UTF8 conversion somewhere.

[giordano]

Should use wpath on Windows?

Edit: nevermind, my understanding is that we want to print the utf8 string and wpath is the utf16 one. So path somehow is utf16?

[apaz-cli]

It should be utf8. The function expects it to be utf8. That it isn't is very odd, and this PR made no changes that would break that.

[giordano]

Yes, but the tests introduced here triggered the error, tests were already failing in this PR, but the fact that Windows tests are allowed to fail means that the failure was overlooked

[inkydragon]

jl_loader_print_stderr3
-> jl_loader_print_stderr
-> fputs
-> loader_fputs
-> utf8_to_wchar + fwrite

julia/cli/loader_win_utils.c

Lines 27 to 31 in b01fd33

	int loader_fputs(const char *str, FILE *out) {
	wchar_t wstr[1024];
	utf8_to_wchar(str, wstr, 1024);
	return fwrite(wstr, wcslen(wstr), out);
	}

So on windows the default error message we output is in wchar.
Maybe it should be converted to UTF8 when reading.
Or we could just output UTF8 without conversion, would that break anything?

julia/test/compiler/codegen.jl

Lines 707 to 710 in b01fd33

	errfile = joinpath(pfx, "stderr.txt")
	@test libs_emptied > 0
	@test_throws ProcessFailedException run(pipeline(`$pfx/bin/$(Base.julia_exename()) -e 'print("This should fail!\n")'`; stderr=errfile))
	@test contains(readline(errfile), "ERROR: Unable to load dependent library")

[apaz-cli]

According to my research, windows can display ascii or utf16 output in terminals, and as long as a single write call is used, it shouldn't matter which.

It seems like the test should detect if it's on Windows and read as utf16 instead?

[PallHaraldsson]

Should this be reverted since causing CI errors elsewhere?