On Jun 22, 2013 23:33, Matthias Kilian wrote:
> On Sat, Jun 22, 2013 at 11:07:25PM +0200, Matthias Kilian wrote:
> > I also wonder if it wouldn't be better to go e step further and use
> > pg_config instead of krb5-config. That one could be submitted to
> > upstream.
>
> I just noticed that the Setup.hs of hs-postgresql-libpq already
> checks for and uses pg_config -- it just doesn't ask it for the
> libraries to link with. So it may be enough to extend psqlBuildInfo in
> Setup.hs with something like this:
>
> ...
> libs <- pgconfig ["--libs"]
>
> return emptyBuildInfo {
> extraLibDirs = [strip libDir],
> includeDirs = [strip incDir],
> extraLibs = map (drop 2) $ words libs
> }
>
> Ciao,
> Kili
>
I played a bit with extending Setup.hs. In general it works fine to set the
extraLibs variable there but unfortunately we can't just use the output of
pg_config or krb5-config because ghci needs them in a certain order.
crypto needs to be in front of ssl.
asn1 needs to be in front of krb5.
If my understanding is right, we need to provide the library names, that
libpq.so depends on so the ghci linker resolves them.
$ ldd /usr/local/lib/libpq.so.5.5 on my system lists the following:
libpq.so.5.5
libkrb5.so.19.0
libasn1.so.19.0
libcrypto.so.22.0
libwind.so.0.0
libroken.so.0.0
libcom_err.so.19.0
libheimbase.so.0.0
libssl.so.19.0
$ pg_config --libs gives:
-lpgport -lxml2 -lssl -lcrypto -lz -lreadline -ltermcap -lm
and $ krb5-config --libs gives
-L/usr/lib -lkrb5 -lasn1 -lcrypto -lwind -lroken -lcom_err -lheimbase
The order and output of all three tools isn't right for the ghci linker.
So we would need to write code that reorders the output and maybe drops some of
the superfluous names.
I think maintaining that code to make sure the right order is maintained is
likely more work in case some Kerberos libs change again compared to the
hardcoded list in the cabal file. What do you think?
Maybe I'm just not seeing the obvious graceful solution :)
