Should we export `get_self_path` and friends?

[LiberalArtist]

In the "Inside" manual, we document ways to embed files (particularly bootfiles) as executable segments or equivalent on various platforms, including finding the path to the running executable:

racket/pkgs/racket-doc/scribblings/inside/appendix.scrbl

Lines 131 to 137 in 8743229

	@section[#:tag "segment-ideas"]{Embedding Files in Executable Sections}
	Locating external files on startup, such as the boot files needed for
	Racket CS, can be troublesome. An alternative to having separate files
	is to embed the files in an ELF or Mach-O executable as data segments
	or in a Windows executable as a resource. Embedding files in that way
	requires using OS-specific linking steps and runtime libraries.

(Rendered at https://docs.racket-lang.org/inside/segment-ideas.html)

Racket also implements these ideas for its own use:

racket/racket/src/start/self_exe.inc

Lines 3 to 14 in 8743229

	Defines self_exe_t get_self_path(char *exec_file),
	which takes argv[0] and returns an improved representation of
	the containing executable. At worst, on Unix, uses `PATH` to
	convert `exec_file` into a path.
	On Mac (even XonX), `find_mach_o_segment` is also defined.
	On Unix (not Mac OS, not even XonX), `find_elf_section`
	is also defined.
	If USE_EXE_LOOKUP_VIA_PATH, also: `lookup_exe_via_path`,
	`path_append`, and `do_path_append`.

Would it make sense for Racket to export these functions in racketcs.h (and scheme.h for BC)?

Someone in another context asked today about embedding files in executables and finding the path to the running program, and I pointed them to what Racket does. The implementations in the "Inside" manual for Windows, Mac, and Linux are reasonably concise, but I remembered there are more issues on the BSDs and other Unix-like systems without /proc/self/exe, and Racket handles these in self_exe.inc.

[LiberalArtist]

Someone on IRC told me that:

the freebsd version is ... suboptimal
for why, see https://github.com/llvm/llvm-project/blob/38f34e587d10fcd7d18fd240e41248006faa639e/llvm/lib/Support/Unix/Path.inc#L209-L214
(the reason why is that KERN_PROC_PATHNAME is obtained by looking up the process text vnode in the namecache, which (a) might not return anything, and (b) will return a name for the executable, not necessarily the one actually used

I have no BSD experience myself, but using the ELF auxiliary vectors sounds like good advice.

They also pointed out that, on Linux, /proc/self/exe "does break if /proc is not mounted within the applicable root or namespace". Currently, Racket calls exit(1) in that case:

racket/racket/src/start/self_exe.inc

Lines 42 to 43 in 8743229

	fprintf(stderr, "failed to get self (%d)\n", errno);
	exit(1);

It might be better to fall back.

[mflatt]

I've made these changes, except for exporting from BC (which doesn't seem worth the extra effort).