On Tue, Oct 27, 2020 at 12:41 AM Alvaro Herrera <alvhe...@alvh.no-ip.org> wrote:
> On 2020-Oct-26, Craig Ringer wrote: > > > Patch 0001 adds PQlibInfo(), which returns an array of key/value > > description items reporting on configuration like the full version > string, > > SSL support, gssapi support, thread safety, default port and default unix > > socket path. This is for application use and application diagnostics. It > > also adds PQlibInfoPrint() which dumps PQlibInfo() keys/values to stdout. > > See the commit message in patch 0001 for details. > > Sounds useful. I'd have PQlibInfoPrint(FILE *) instead, so you can pass > stdout or whichever fd you want. > The decision not to do so was deliberate. On any platform where a shared library could be linked to a different C runtime library than the main executable or other libraries it is not safe to pass a FILE*. This is most common on Windows. I figured it's just a trivial wrapper anyway, so people can just write or copy it if they really care. > Patch 0002 exposes LIBPQ_VERSION_STR, LIBPQ_VERSION_NUM and > > LIBPQ_CONFIGURE_ARGS symbols in the dynamic symbol table. These can be > > accessed by a debugger even when the library cannot be loaded or > executed, > > and unlike macros are available even in a stripped executable. So they > can > > be used to identify a libpq binary found in the wild. Their storage is > > shared with PQlibInfo()'s static data, so they only cost three symbol > table > > entries. > > Interesting. Is this real-world useful? I'm thinking most of the time > I'd just run the library, but maybe you know of cases where that doesn't > work? > It was prompted by a support conversation about how to identify a libpq. So I'd say yes. In that case the eventual approach used was to use Python's ctypes to dynamically load libpq then call PQlibVersion(). > Patch 0003 allows libpq.so to be executed directly from the command line > to > > print its version, configure arguments etc exactly as PQlibInfoPrint() > > would output them. This is only enabled on x64 linux for now but can be > > extended to other targets quite simply. > > +1 --- to me this is the bit that would be most useful, I expect. > It's also kinda cool. But it's using a bit of a platform quirk that's not supported by the toolchain as well as I'd really like - annoyingly, when you pass a --entrypoint to GNU ld or to LLVM's ld.lld, it should really emit the default .interp section to point to /bin/ld.so.2 or /lib64/ld-linux-x86-64.so.2 as appropriate. But when building -shared they don't seem to want to, nor do they expose a sensible macro that lets you get the default string yourself. So I thought there was a moderate to high chance that this patch would trip someone's "yuck" meter.
