[bitc-dev] bug in character decode routines

[Sandro Magi]

Jonathan S. Shapiro wrote:
> I know C-- was rejected as a backend; I wonder if LLVM was evaluated:
> 
> http://llvm.org/
> 
> It even provides an LLVM IR to C translator.
> 
> Yes, we looked. Decided the learning curve would take too long.

I don't want to beat on what is now a moot point, but surely learning a
compiler framework that has has many man-years of effort put into it is
easier than building one of your own and likely stumbling over the
problems said framework already solves?

> In particular, there were complexities in dealing with GC, and we didn't want to lose the time to figure them out.

By complexities, do you mean the co-ordination required between the
front-end and the backend to find all pointers? This didn't seem all
that complicated when I looked into the LLVM GC a couple of months ago.

I knew hardly anything about GC at the time, but I managed to write a
ref-counting GC for LLVM
(http://lists.cs.uiuc.edu/pipermail/llvmdev/2006-March/005337.html),
after a few questions with the maintainers (who are very responsive to
inquiries).

> LLVM really wants you to have an optimzer. We didn't ant to have to build that for a prototype system.

I was under the impression that optimization was a completely separate
phase in the LLVM pipeline. Sorry for all the questions, but I was
thinking of using LLVM in a project, and I'm interested in your conclusions.

Sandro