Re: Bug#1143030: netkit-ntalk: FTBFS on hurd-amd64: PATH_MAX and MAXHOSTNAMELEN undeclared

Hi,

I guess thanks for fixing Hurd portability bugs. Ideally these fixes 
would go upstream instead of just into Debian.

* Federico Bonino <[email protected]> [260729 22:35]:
>netkit-ntalk fails to build from source on hurd-amd64 because two macros
>the source relies on are not defined by the Hurd libc headers:

>Buildd log (hurd-amd64, 0.17-20):
>https://buildd.debian.org/status/fetch.php?pkg=netkit-ntalk&arch=hurd-amd64&ver=0.17-20&stamp=1783711377&raw=1
>
>  talk/get_names.c:55:23: error: 'MAXHOSTNAMELEN' undeclared (first use in
>this function)
>     55 |         char hostname[MAXHOSTNAMELEN];
>
>The build stops there. One file later, talkd/process.c:122 declares
>find_user()'s scratch buffer as "char besttty[PATH_MAX]", which does not
>compile on Hurd either.
>
>Root cause: GNU/Hurd has no fixed path limit, so PATH_MAX is not defined.
>MAXHOSTNAMELEN is a BSD constant that is also absent on Hurd.
>
>The patch below replaces the PATH_MAX-sized temporary buffer in
>talkd/process.c with dynamic allocation. It also supplies MAXHOSTNAMELEN
>from HOST_NAME_MAX, or from _POSIX_HOST_NAME_MAX when HOST_NAME_MAX is
>unavailable. No feature or test is disabled.
>
>I verified the patch on native GNU/Hurd amd64 running Debian unstable. The
>unmodified 0.17-20 source reproduced the MAXHOSTNAMELEN build failure. The
>patch then applied with zero fuzz and zero offset, and dpkg-buildpackage
>completed successfully, producing talk and talkd. The package defines no
>autopkgtest or upstream test suite.
>
>One thing to flag rather than let you find it in your own build log: on GCC
>15 the new strdup(uptr->ut_line) call raises
>
>  warning: 'strdup' argument 1 declared attribute 'nonstring'
>[-Wstringop-overread]
>
>because ut_line is a fixed-size utmp field with no guaranteed NUL
>terminator.
>The strcpy() it replaces read ut_line exactly the same way, so this is a
>new diagnostic rather than a new unbounded read — but it is new, and it
>appears on every architecture. If you would prefer the variant without it,
>sizing besttty
>as sizeof(uptr->ut_line) + 1 and copying with an explicit terminator
>removes the PATH_MAX dependency just as well and is a smaller diff. I am
>happy to send that instead; I went with the allocation because it makes no
>assumption about the utmp field width.
>
>Toolchain stamp:
>- hurd 1:0.9.git20260527-3+b1
>- libc0.3 2.42-17
>- gcc-15 15.3.0-1
>- binutils 2.46.90.20260712-1
>
>netkit-ntalk is maintained by the Debian QA Group and the netkit collection
>has no active upstream, so there is nowhere to forward this fix — hence
>"Forwarded: no" in the DEP-3 header below, and this report rather than an
>upstream submission.
>
>Assisted-by: hurd-collab agent (automated portability triage and fix)

Could you take some time to explain what "hurd-collab agent" is 
exactly? I could not find anything in a quick web search.

Best,
Chris

>Signed-off-by: Federico Bonino <[email protected]>
>
>--- netkit-ntalk-hurd.patch ---
>Description: Replace PATH_MAX buffer with dynamic allocation and add
>MAXHOSTNAMELEN fallback for GNU/Hurd
>Source-Version: 0.17-20
>Author: Federico Bonino <[email protected]>
>Forwarded: no
>Last-Update: 2026-07-28
>
>GNU/Hurd does not define PATH_MAX because it has no fixed path limit. The
>besttty buffer in talkd/process.c was only used as a temporary scratch area
>to hold a utmp tty line before copying it to the caller's output buffer; it
>is now dynamically allocated with strdup, removing the fixed-size PATH_MAX
>dependency.
>
>The MAXHOSTNAMELEN buffer in talk/get_names.c is used for gethostname and
>the legacy talk wire protocol uses fixed-size fields, so a bounded hostname
>buffer remains appropriate. Since Hurd does not provide MAXHOSTNAMELEN,
>provide a conditional fallback to HOST_NAME_MAX (or _POSIX_HOST_NAME_MAX)
>from
><limits.h>, preserving the existing behaviour on BSD/Linux platforms.
>
>--- a/talkd/process.c
>+++ b/talkd/process.c
>@@ -55,6 +55,7 @@ char rcsid[] =
> #include <syslog.h>
> #include <stdio.h>
> #include <string.h>
>+#include <stdlib.h>
> /* #include <paths.h> <---- unused? */
> #include <utmp.h>
> #include "prot_talkd.h"
>@@ -119,9 +120,12 @@ find_user(const char *name, char *tty)
>  struct utmp *uptr;
>  int found=0, ok=0, ret;
>  time_t best_time = 0, this_time;
>- char besttty[PATH_MAX];
>+ char *besttty;
>
>- *besttty = 0;
>+ besttty = strdup("");
>+ if (besttty == NULL) {
>+ return FAILED;
>+ }
>  setutent();
>  while ((uptr = getutent())!=NULL) {
> #ifdef USER_PROCESS
>@@ -135,6 +139,7 @@ find_user(const char *name, char *tty)
>  if (*tty && !strcmp(tty, uptr->ut_line)) {
>  /* asked for a tty, found it */
>  endutent();
>+ free(besttty);
>  return SUCCESS;
>  }
>  ret = check_tty_perms(uptr->ut_line, &this_time);
>@@ -145,11 +150,17 @@ find_user(const char *name, char *tty)
>  found = ok = 1;
>  if (this_time > best_time) {
>  best_time = this_time;
>- strcpy(besttty, uptr->ut_line);
>+ free(besttty);
>+ besttty = strdup(uptr->ut_line);
>+ if (besttty == NULL) {
>+ endutent();
>+ return FAILED;
>+ }
>  }
>  }
>  endutent();
>  strcpy(tty, besttty);
>+ free(besttty);
>  return !found ? NOT_HERE : (!ok ? PERMISSION_DENIED : SUCCESS);
> }
>
>--- a/talk/get_names.c
>+++ b/talk/get_names.c
>@@ -39,6 +39,14 @@ char gn_rcsid[] =
>
> #include <stdio.h>
> #include <sys/param.h>
>+#include <limits.h>
>+#ifndef MAXHOSTNAMELEN
>+#ifdef HOST_NAME_MAX
>+#define MAXHOSTNAMELEN HOST_NAME_MAX
>+#else
>+#define MAXHOSTNAMELEN _POSIX_HOST_NAME_MAX
>+#endif
>+#endif
> #include <sys/socket.h>
> #include <pwd.h>
> #include <unistd.h>