Gem fails to compile

[dnebdal]

Hi. Using ruby 2.7.2 on Fedora 33, both "gem install lmdb" and a git clone / bundle exec rake fails with very similar errors:

gcc -I. -I/usr/include -I/usr/include/ruby/backward -I/usr/include -I../../../../ext/lmdb_ext -I../../../../ext/lmdb_ext/liblmdb -DRUBY_EXTCONF_H=\"extconf.h\"    -fPIC -std=c99 -Wall -g -m64 -o lmdb_ext.o -c ../../../../ext/lmdb_ext/lmdb_ext.c
../../../../ext/lmdb_ext/lmdb_ext.c: In function ‘call_with_transaction_helper’:
../../../../ext/lmdb_ext/lmdb_ext.c:172:10: error: #error "Not implemented"
  172 |         #error "Not implemented"
      |          ^~~~~

As far as I can tell, this is this block in lmdb_ext.c , starting at line 169:

// Ruby 1.8.7 compatibility
#ifndef HAVE_RB_FUNCALL_PASSING_BLOCK
static VALUE call_with_transaction_helper(VALUE arg) {
        #error "Not implemented"
}
#else
static VALUE call_with_transaction_helper(VALUE arg) {
        HelperArgs* a = (HelperArgs*)arg;
        return rb_funcall_passing_block(a->self, rb_intern(a->name), a->argc, a->argv);
}
#endif

Just removing the entire ifdef and uncritically using the #else version seems to work for me, but I assume it's there for a reason.

[doriantaylor]

hmmm I'm gonna assume that's reeeally old. One moment, I haven't looked at this in a while.

[doriantaylor]

I just successfully compiled on 2.7.0 on Ubuntu 20.04 and 3.0.1 on macOS. It looks like your extconf.rb is not picking the presence of rb_funcall_passing_block. (or not being run at all?)

Is it possible to see what your extconf.h looks like, the one that gets generated when you try to bundle exec rake?

Also since I've taken over maintenance of this gem, can you try building my fork?

[dnebdal]

Same thing from your fork, I'm afraid. It's entirely possible that this is on me; I very rarely use Ruby and I'm going by guesswork and google for how to build gems.

Here's a log of what I did, how it failed, and what's in extconf.rb and extconf.h :
https://gist.github.com/dnebdal/1683fdbcc4deb777950aa5b8ba3a96f1

[doriantaylor]

hm, as predicted those defines are missing. eeeeeenteresting.

[dnebdal]

I can try to recreate it in a Fedora VM, just to rule out any weird local variations?

[doriantaylor]

If extconf.rb isn't setting those defines that probably means it can't see the header files, but if you can compile it otherwise it must necessarily be able to see them, so that is really weird. Can you double-check that you have the dev headers for ruby? Also is it possible that your environment variables are getting messed with?

[doriantaylor]

actually, what is -I/usr/include/ruby/backward ? are there old headers in there masking the new ones?

[doriantaylor]

Got somebody on Freenode to reproduce on Fedora with 3.0.1; inclined to chalk this up as a Fedora problem. Something is messing with the execution of that extconf.rb.

[fnordfish]

For me, doriantaylor@67e7d1e did the trick for a docker rockylinux:8 install