Re: error-tokens and locations?

[Ian Badcoe]

At 11:16 14/11/2003 -0800, you wrote:
> Ian Badcoe <address@hidden> writes:
>
> >          I could grind the rust of my gdb skills and look into the crash.
> >          Or, I look into the build some more and see if I'm doing it wrong.
> >          Or, I could just post the problem demo.
> >          What would you prefer?
>
> The first sounds easiest for all concerned.

Hmm, there's quite a lot of rust on my gdb, maybe 10 years worth . . . :)

I've had a quick look, but at first glance I cannot even see where to start 
with configure (or the non-human-friendly makefiles it produces) to ask it 
for a debug version of the build.

I will have another look on another day, but I don't have a lot of time and 
I can manage without perfect error messages for the moment.  I'd only be 
able to find out why the build goes wrong anyway, it wouldn't tell me about 
the bug...

I did demonstrate that I can build and run the 1.875b sources which came 
from cygwin, so my build environment is working.  I don't know whether the 
cygwin people patch it at all.

I'm attaching some files to demonstrate the problem, just in case anybody 
wants to see.  I have built them with 1.875c under cygwin and they have the 
same problem as my previous builds, with the minor exception that the 
uninitialised locations have zeros in them under gcc, where they had random 
bit-patterns under VC.

> Another possibility is that you could build 1.875c on GNU/Linux and
> try your problem demo there.

If I had such a machine, it would be the first thing to try.  I did take a 
quick look around the web for GNU/Linux installs, but there's all sorts of 
issues of partitioning and multibooting and the size of the downloads etc etc.

The attached zip has 4 source files and a very very basic makefile ("make 
all").

temp.txt is the output I got when I ran the demo program under cygwin.  The 
problem is illustrated by the first "MERGING LOCATIONS" printf, where the 
number in square-brackets is the number of tokens being merged (it should 
be base-1 indexed, so zero is a very bad value).

make check.txt is the result of running make check on the bison 1.875c 
build.  It seems pretty unhappy.

        thanks,
                Ian B

p.s. I will diff the 1.875b (cygwin) and 1.875c sources.  If you could 
point me at a definitely non-patched 1.875b then that would be simpler to 
compare...

temp.zip
Description: Zip archive

Free transport into the future, while you wait.