[Bug-zile] astr_cstr

[Reuben Thomas]

Re. a recent commit:

commit 5171aa758f7354e8db522f341a2dc55563e82a62
Author: Gary V. Vaughan <address@hidden>
Date:   Thu Sep 1 21:57:27 2011 +0700

    compiler warning: assignment to as->text discards const

    remove the constness of the string passed to castr_new_nstr, because
    once its address is held in as->text, the rest of the code will not
    treat it as const anyway; thus, we also have to remove the constness of
    return type of astr_cstr() which is simply giving the address in the
    non-const as->text pointer too.

I am unhappy with this, because it removes compiler warnings about two
sorts of potential bug:

1. Code overwrites the return value of astr_cstr and the compiler
doesn't complain. (astr_cstr returns a read-only copy of an astr as a
char *; it's pure coincidence that it happens to be the actual
representation too.)

2. Code passes a non-const string to castr_new_nstr. The whole point
of this constructor is that the string really is const! (Look at the
code for castr_new_nstr). The underlying problem here is trying to
have two structured types, astr and castr, which are compatible but
one of which is immutable.

This is a good example of where trying to keep the compiler happy
actually makes things worse (and why the GCS does not insist on
warning-free compilation, even with the GNU compiler).

-- 
http://rrt.sc3d.org