Robert Schwebel <[EMAIL PROTECTED]> writes: > Configuring on linux-2.4.9-rthal6 with glibc-2.2.2 without switches (just > added paths to headers and lib of gmp) works fine in the beginning, > besides lots of warnings like this one: > > atoms_gperf.c:332: warning: missing initializer > atoms_gperf.c:332: warning: (near initialization for `wordlist[487].id')
These are harmless. The code is automatically generated, which is why I don't think it's very impotant to fix these warnings. > This seems to come from several $(addprefix ../,) commands in the makefile > which obviously are missing their second parameters. Seems to be a problem > in the Makefile.am, but I simply removed the lines in the Makefile and it > worked afterwards. This seems to be a bug in the Makefile.am. It reads LDADD = ../liblsh.a ../symmetric/libsymmetric.a $(addprefix ../,@LIBARGP@) which breaks if LIBARGP is empty. I can't see right away what the right way is to fix it. If lsh's own argp is used, then it should work, because configure should set LIBARGP to "argp/libargp.a". > Now, the more difficult part ;-) My goal is to run lsh on an embedded > machine with very restricted ressources, so I'm currently trying to > compile and link it against uClibc. One thing that you may have to consider once the compilation problems are resolved, is sources of randomness. The code in unix_randomness.c may not get enough entropy on an embedded system, and that may break any security in lsh. Or if you still have a good /dev/urandom, that might be good enough. > Here's where I'm stuck at the moment: uClibc doesn't have argp_parse, > which was recognized correctly by the configure script and therefore lsh's > own implementation is used. The linker breaks up here: > Very strange, as it was weak_aliased() in argp-parse.c: > > #ifdef weak_alias > weak_alias (__argp_parse, argp_parse) > #endif > > The original glibc implementation looks like this: > > robert@callisto:~/embedded/glibc-2.2.4/argp ! \ > nm -n argp-parse.o |grep argp_parse > 00000e40 T __argp_parse > 00000e40 W argp_parse > > Which seems to be correct. All this weak alias stuff and symbols with a "__" prefix (which I don't fully understand) should only be used when argp is compiled as a part of glibc, not when it is compiled stand-alone like with lsh. Look at argp-namefrob.h. If _LIBC is undefined (like in lsh's case), it just #defines __argp_parse argp_parse. Regards, /Niels
