laman@sdcsvax.UUCP (Mike Laman) (08/13/83)
I don't think this got off this system, so I am resubmitting it.
There is a line in the System III source of "ed", that is like:
p = mktemp("/tmp/edXXXXXX");
Now as you may know "mktemp()" writes on the given argument. I came
across a compiler that mapped strings into the linkage section.
Obviously, when "mktemp()" tried to write on the string, the program
died since the linkage section was write protected. My question is
this:
What does "The C Programming Language" say about this problem? Was
the compiler incorrect according to Kerningham and Ritchie?
I am NOT asking if people SHOULD do it or not. That will be as
interesting a discussion as where people put their curly braces.
I just want to find what is correct. You might as well submit it to the net;
this could be interesting. I can't find the answer in the book.
Mike Laman
UUCP: {ucbvax,philabs,sdccsu3}!sdcsvax!laman
ARPA: mike@NPRDCjohnl@ima.UUCP (08/17/83)
#R:sdcsvax:-364800:ima:15900010:000:760 ima!johnl Aug 16 12:18:00 1983 The question is asked whether a C compiler that makes strings read-only is in conformance with K&R.. It appears that it is indeed wrong. The C reference in section 2.5 states that strings are semantically static arrays of characters. The discussion of "static" in section 4 talks about variables, not constants. You might read the exact words a little differently, but the intention is pretty clear. In practice, zillions of existing C programs expect to be able to write on their strings, so that read-only strings will break programs regardless of the moral status of the compiler. John Levine, ima!johnl PS: There has been some discussion of a "readonly" keyword to mark data that should be read-only, but I never heard of it being blessed at Bell.
jmc@root44.UUCP (John Collins) (08/17/83)
Personally, I think that constants in a program ought to be exactly that, and not something which might be what they look like the first time through, and something else the next time. There is an old chestnut about assigning to a Fortran argument which has been passed as a constant. Some languages (e.g. BCPL, and (in a different sense) COBOL) allow you to change 'goto's whilst the program is running to make them 'goto' somewhere else. UGH!! However K&R p181 says that strings are 'storage class static'. I suppose that does decide the issue. I think we do need a syntax for 'readonly' storage classes, both for strings and external/static stuff so we can stop patching assembler output, however that subject has already been aired rather a lot... John Collins ....!vax135!ukc!root44!jmc
dalka@ihuxe.UUCP (08/17/83)
The "C Programming Language" book leaves many issues untouched. This happens to be one of them. So essentially its up to the compiler implementor as to what to do. This is why "C" is not as portable as people make it out to be...... Ken Dalka ihuxe!dalka
msm@syteka.UUCP (Michael S. Maiten) (09/07/83)
The iNTEL 186 C compiler (really Mark Williams C) has an option specified
when invoking the compiler that will put ALL strings into the constant segment
(which can be a section of ROM addressed by the 186 data-segment register).
Michael S. Maiten
Silicon Gulch, California
<...!{ucbvax!menlo70,decvax}!sytek!msm>