Segfault while bootstraping from ruby

Help & Support
5

Hi, @GeopJr and other readers…

I’ve also tried to get the bootstrap script working and ran into similar issues (on Debian 12 “bookworm”, in my case). Here’s what I did so far:

  1. Like you, I saw I needed Python 2. I built Python 2.7.18 from source and made a symlink so that the python2 command would run this version.
  2. I saw a message about needing to install GraphViz. This is a red herring: GraphViz was already installed, and this message is displayed regardless of whether or not it’s installed. So, you don’t need to worry about this message.
  3. I had an error about “Cannot find llvm-config”. This was displayed immediately after the GraphViz message (which is why I ended up looking into that). I’m still not sure why this happened, but after a bit of investigating, I ended up just doing rm -rf buildroot/src/llvm-3.3.src and re-running the bootstrap script, and it worked that time.
  4. I got the error with librt.so. Now I can share some useful information about this error: it happens because that library has been removed from modern glibc. It was only needed by time.linux.cr, which is easily fixed with the following patch (add this to the end of stage1.patch, and note that the empty-looking line before class Time needs a single space on it to be a valid patch file):
diff --git a/std/time.linux.cr b/std/time.linux.cr
index 8aa07734b..dbe094b9a 100644
--- a/std/time.linux.cr
+++ b/std/time.linux.cr
@@ -1,4 +1,4 @@
-lib Librt("rt")
+lib C
   struct TimeSpec
     tv_sec, tv_nsec : Int64
   end
@@ -7,7 +7,7 @@ end
 
 class Time
   def initialize
-    Librt.clock_gettime(0, out time)
+    C.clock_gettime(0, out time)
     @seconds = time.tv_sec + time.tv_nsec / 1e9
   end
 end

If you search for clock_gettime, you might find yourself [here]( https:// www . man7 . org /linux/man-pages/man3/clock_gettime.3.html ), which states “Link with -lrt (only for glibc versions before 2.17)”; that’s how I figured to just change Librt to C.

  1. I got an error about libunwind.so. I tried the following workaround (I don’t suggest doing this, as it doesn’t seem to work, see below):
# as root
cd /usr/lib/x86_64-linux-gnu
ln -s libunwind.so.8.0.1 libunwind.so

Doing this does allow stage1 to build, but when it tries to actually run it, it gets a segfault - maybe the same segfault that you’re getting…

I did do a little research, which reveals there’s multiple different libunwind projects, and maybe we’re linking to the wrong one. My libunwind.so.8.0.1, which is part of my Debian 12 system, comes from [“nongnu” libunwind]( http:// www . nongnu . org /libunwind ), whatever “nongnu” means. But there’s an LLVM libunwind as well, as discussed [here]( https:// github . com /libunwind/libunwind/discussions/665 ), and I see it’s mentioned [here]( https:// clang . llvm . org /docs/Toolchain.html#libunwind-llvm ) in the LLVM docs. Since Crystal uses LLVM I’m pondering whether I need to somehow get it to link with the LLVM libunwind instead of the “nongnu” one, and whether that’ll resolve the segfault.

But I’m getting very out of my depth here, so that’s where I stopped. If anyone has some pointers for how to make further progress, I’d really appreciate it!

Hopefully related, I’ve also posted [over on the Debian User Forums]( https:// forums . debian . net /viewtopic.php?t=160984 ) to try to find out how Debian is compiling Crystal, since I feel sure they must be bootstrapping it somehow.

PS. This “new users can only put two links in a post” rule isn’t very nice! I’ve bloated all the links with spaces to get round this, you’re just going to have to manually de-space them.

1 Like