Hi Mark, On Wed, Oct 26, 2022 at 11:06 AM Mark Wielaard <m...@klomp.org> wrote: > On Mon, 2022-10-24 at 14:38 -0400, Frank Ch. Eigler via Elfutils-devel > wrote: > > - not sure I understand why the code worries about dots in or not in > > section names. Why not just pass them verbatim throughout the code > > base, and not worry about whether or not there's a dot? Does the > > ELF standard even require a dot? > > I agree that just passing them through as is might be better. The ELF > standard doesn't say much about section names, just: > > Section names with a dot (.) prefix are reserved for the system, > although applications may use these sections if their existing > meanings are satisfactory. Applications may use names without the > prefix to avoid conflicts with system sections.
Agreed, will fix. > I would drop the maybe_debuginfo_section heuristics. There are some > sections like .strtab/.symtab that are probably in the debug file, but > might be in the executable. I would assume that a named section can > normally be found in the debugfile and only use the executable as > fallback. > > So see if you can find the .debug file, if you can, then look for the > section by name. If it isn't SHT_NOBITS you found it. If it is > SHT_NOBITS the section should be in the exe. If the section cannot be > found by name (in the .debug file) you can stop searching, it also > won't be in the exe. If you cannot find the .debug file, or the section > was in the .debug file, but had type SHT_NOBITS then search for the exe > file and the named section in there. I like this heuristic. It's simpler and we don't have to update anything if/when a new section becomes common. > Finally, if the section comes from a file in the cache or if we have to > download it in full anyway, then extracting the section into its own > file seems slightly wasteful. It would be great if we could just report > back "here is the full exe/debug file which does contain the requested > section name". But that might make the interface a little ugly. > > int > debuginfod_find_section (debuginfod_client *client, > const unsigned char *build_id, > int build_id_len, > const char *section, char **path, > bool *file_is_elf) > > Maybe that is over-designed to avoid a little bit of disk waste? We'd save a bit of disk space but complicate the API and often cause client tools to have to do more work to get at the section contents. In the case of .gdb_index, gdb already knows how to read the index from a separate file. Of course it can read the section from an ELF file too but I suspect there might need to be some changes to teach it how to handle "unusual" ELF files that only contain a single section. > Since debuginfod-client.c already includes system.h it can use: > > static inline ssize_t > write_retry (int fd, const void *buf, size_t len) > > Which takes care of partial and/or interrupted write calls. Thanks that's exactly what I'm looking for. Aaron
