LLVMdev Digest, Vol 61, Issue 25

Hi Anton,

The problem is in getting the system configured. Below is the crash I get in configure. Your comment regarding gnuwin32 not being needed is interesting - I actually installed gnu bison/flex to get past an earlier crash in configure.

Thats why I was suggesting that a precompiled OBJ_ROOT for MingW32 would be a good idea - configure appears to be inherently fragile, and requires more expertise than I have to get past it if it does fail.

Configure looking like a fragile thing, and windows programmers (like myself) lacking unix expertise to work arround it if it fails - the configure script is 1MB in size - 36000+ lines - which is pretty daunting.

I am getting the exact same crash for the configure script for libjit as well by the way, so I know it's my fault. But I'm not sure whats broken. Do I have a bad version of 'sed' or something?

I am already using MinGW for my own work, so my basic configuration works well enough. I installed msys1.0 so I could build llvm, but have no other need for it. If I could get at a pre-built OBJ_ROOT, I still wouldn't need it I believe.

NB: The configure script for libjit is 800K

Hello, Lance

Your comment regarding gnuwin32 not being needed is interesting -
I actually installed gnu bison/flex to get past an earlier crash in

Which version of LLVM you're building?


mingw32 gcc 4.x should be considered as 'alpha'. 3.4.5 is known to be stable.

This might give you the needed information about LLVM build on mingw32:

Some stuff listed there is not needed anymore (flex / bison) and some
versions are definitely not up to date, but otherwise it's pretty

LOL, Hi Anton,

LLVM-2.5 - should have been the first thing I said. Sorry bout that.

I had actually read that www site, but hadn't recognised your name. Pretty much duplicates my own experience. Anyway, I haven't followed the instructions through carefully, as I am finding it difficult to locate the downloads listed - from march 2008. MinGW doesn't seem to have a single coherent point of download as far as I can tell.

mingw gcc version as you ask is 4.3.3 - but I have the llvm-gcc compilers to avoid any issue there if/when I get that far. A couple of hours ago it was 3.4.5 - I ended up with 4.3.3 when I used an msys1.0 image I downloaded somewhere.

I will follow that site - excluding the last three "tad tricky" ones - bison/flex/regex - which you say should not be needed. Actually those three look like the workarround I needed to get arround the first crash I saw in configure.

When I get this building I can zip up the resultant OBJ_ROOT and pass it back if llvm.org has somewhere to host it - I add that I intend to use llvm-gcc to build which should suit. Interested? :slight_smile:

Hello, Lance

10:46:08.20 C: mingw32-make
gcc -IC:/MinGW/include -g -c que.cc
que.cc:1:19: stdio.h: No such file or directory
que.cc:2:20: stdlib.h: No such file or directory
que.cc:3:20: string.h: No such file or directory

Have you really unpacked mingw-runtime and w32api tarballs? It seems you didn't.

Hello, Lance

With regards to my original problem, still requires flex/bison to configure,
and I have installed. The problem that brought me to the list turns out to
be that I have a copy of borland/inprises grep tool on the path (which I use
a lot).

Heh, name clashing is well-known problem. Another source of such
problems is 'sort' util.

Probably not an issue, but I see in the coding standards chris likes
warnings to be treated like errors. Is there something else I am missing?
Perhaps an env setting to disable those warnings?

The mentioned stl_uninitialized warnings are harmless and comes from
compiler bug. Unfortunately, there is no way to "turn them off".