Re: more ctype(3) UB has crept in

[James Browning]

On Oct 31, 2026 12:40, "Gary E. Miller" <g...@rellim.com> wrote:

Yo Greg!

On Sat, 31 Oct 2026 08:11:27 -0400
Greg Troxel <g...@lexort.com> wrote:

> I thought we had fixed all of these, but I'm building/testing git
> master (to verify that WRITE_PAD is still needed), and see the
> following warnings in the build.   This is the signature of invoking
> a ctype(3) function with type char, which could be signed, and if
> negative, would be promoted to a negative int, which gets you into
> UB, per ISO C -- the only allowable values are those representable as
> unsigned char and EOF.

The POSIX 2024 man page is here:

https://pubs.opengroup.org/onlinepubs/9799919799/functions/iscntrl_l.html

int iscntrl(int c);

Been that way since POSIX.1-2008.


From C99:

7.4.1.4 The iscntrl function

    Synopsis
   
    #include <ctype.h>
    int iscntrl(int c);

gpsd requires C99 and POSIX 2008.  So the current code is correct.

My rat brain is screaming at me that it has come up before that someone had gpsd trouble on a platform with unsigned chars. I am too lazy to look up the reference right now.

Ah, I think it was ESR and Fred Wright talking about IBM mainframe stuff that I misremember. 370?