Re: [Liberty-eiffel] producing debug version od compile_to_c

[Raphael Mack]

Hi Laurie,

the install script compiles the compiler with

compile_to_c -verbose -boost -no_gc compile_to_c -o compile_to_c.new

So the difference is the garbage collector, which is switched off  
normally for compile_to_c (and some other tools). And you probably  
know the current situation, that eiffeldoc fails to generate the html  
on http://doc.liberty-eiffel.org
 which is related to a problem in the GC when it seems to free some  
still referenced (probably in a native array) objects (which likely  
yields a segfault) .

Debugging the compiler ist quite hard, as the contracts in it are not  
as good as we'd like them to be and if you check them, it gets so  
slow, that it is impossible to work with it. Often using "-verbose"  
when compiling the target application in combination with reading the  
source compiled before the crash helps to extract the reason for the  
crash into a small examlpe to reproduce the bug. (a Segfault is always  
a bug in the compiler, even if the input is also malicous). I'd be  
interested in this kind of small example code showing the bug, to be  
able to add it to the test suite.

I hope you still want to go for it, so you might want to look at  
Liberty/work/debug/checks/c2c.ace which is a starting point to check  
SOME contracts in the compiler. But with this approach the challenge  
is to actiavte the interesting contracts and disable all which make  
the thing slow ;-)

For segfaults it is probably best to debug compile_to_c with gdb (add  
"-g" to the command line")...

Cheers,
Rapha

Zitat von Laurie Moye <address@hidden>:

> Hi,
>
> I have a program which crashes compile_to_c when I try to compile it  
> with the adler version.
>
> I have tried to create debug versions of compile_to_c to see if I  
> can find out what's happening, but every version  that I create,  
> even a -boost version compiled with:
> compile -boost compile_to_c -o compile_to_c-boost
> produces a compile_to_c that fails with SIGSEGV even when I try to  
> use it to compile hello_world.
>
> I can see that my compile_to_c produced in this way is of a  
> different length to the one produced by he install process, so I'm  
> obviously doing something wrong.
>
> Is there a standard way of producing debug versions of compile_to_c  
> in order to track down failures?
>
> Best wishes,
>         Laurie