While testing my bytecode reader changes, I ran the regression tests and all 
the tests involving lli failed. The reason was that the bytecode reader was 
throwing an exception that lli was not handling so an abort occurred as the 
result of unhandled_exception(). While the bytecode reader was erroneously 
throwing exceptions, it is not inconceivable that it could do so. A disk I/O 
error could cause a mis-read which would almost certainly cause it to fail. 
The exception thrown contains a std::string& that has the error message. 

lli should be improved to catch std::string& from bytecode reader and 
terminate gracefully with the error message.

It doesn't?

The place I see that could throw an exception is in JIT.cpp:117, which does 
catch it.  What am I missing?

-Chris

Wish I'd pasted the stack trace in here now. I'll try to recreate this (hard 
to do) and let you know the details.

thx :)

any luck?

-Chris

If I get the bytecode doc done before bedtime, I'll look at reproducing this
tonight, otherwise tomorrow night.

Created an attachment (id=152) [details]
Proposed patch to make exception handling around calls into bcreader more
robust.

Its difficult to tell if this fixes the reported problem. In order to replicate
the problem, it is necessary to hand modify a bc file in such a way that a
function definition causes the bcreader to throw an exception. Several attempts
to do this failed. The patch is the result of using llvmgrep to find all the
uses of "materialize" and augment the exception handling accordingly.

I bet FunctionPassManager::run is the one that you ran into before.  The patch 
looks good with the addition of an 'abort' call in the second hunk.  With that 
change, please apply. :)

Thanks Reid!

-Chris

Yeah, that's pretty much what I was thinking.

Resolved. Patches here:

http://mail.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20040705/015804.html
http://mail.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20040705/015805.html
http://mail.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20040705/015806.html
http://mail.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20040705/015807.html