[Aldor-l] aldor.conf::gcc-hacks

[Gabriel Dos Reis]

Hi,

  The Aldor compiler, by default, passes the option -fwritable-strings
to GCC, which is unrecognized by GCC-3.4.x and later; therefore
causing the compilation to halt.

  I tried to commment out that flag.  However, there is a comment
in aldor.conf saying:

   # On some platforms string literals are stored in read-only segments.
   # This causes problems when users write Aldor programs in which a
   # string literal is used as a buffer. The result is a segfault that
   # is extremely hard to track down, even for experienced Aldor coders.

I'm doubly concerned: both as an aldor user and a GCC developer/maintainer.

As an aldor user, my concern is that if I remove that option -- a
necessary removal is I ever want to get my aldor programs compile --
then programs may be subject to random misbehaviour that would be hard
to track down.

As a GCC developer/maintainer, I read the coment to say that it might
be a bug in GCC.  But I cannot think of any scenario where that
is the case.  

If the aldor compiler generates code like

    char* text = "string-constant";

and the pointer "text" is used to modify the pointed-to string, then
that is a bug in the program generated by the aldor compiler. 

If the aldor compiler generates code like

    char text[] = "string-constant";

then the object "text" is modifiable and GCC should not put it in
read-only sections.  And if that is not the case, the it is a bug in
GCC; but all data I have suggest that GCC does not have such bug.

Since the aldor souce code is not avaliable for public inspection, I
would have some questions:

   (1) which of the above code generation scenario is the case?
   (2) is it really unwise to remove the -fwritable-strings option, or
       it is only neecessary for some "corner" cases?
   (3) is there a plan to fix that in the next version of aldor?
       (no recent version of GCC will accept the option).

Thanks,

-- Gaby