------- Additional Comments From sbellon at sbellon dot de 2005-02-27 18:47
-------
I have investigated further. There are further places in the C part of the GNAT
library that need changing because they are inconsistent.
In __gnat_portable_spawn in ada/adaint.c the spawn/exec calls are always the
'path' variants except for Unix. The patch for GCC 3.4.3 should be as follows:
--- adaint.c 2004-01-13 12:51:31.000000000 +0100
+++ patched/adaint.c 2005-02-25 13:25:40.000000000 +0100
@@ -1520,9 +1520,9 @@
if (pid == 0)
{
/* The child. */
- if (execv (args[0], args) != 0)
+ if (execvp (args[0], args) != 0)
#if defined (VMS)
- return -1; /* execv is in parent context on VMS. */
+ return -1; /* execvp is in parent context on VMS. */
#else
_exit (1);
#endif
The only excuse could be that not all operating systems support execvp. In that
case, a simple define could fix that situation.
Furthermore, in ada/expect.c there are several versions of
__gnat_expect_portable_execvp that in fact don't call the 'path' variant of the
spawn/exec call although the function is called __gnat_expect_portable_execvp.
The following patch fixes that:
--- expect.c 2003-10-31 02:08:42.000000000 +0100
+++ patched/expect.c 2005-02-25 13:31:57.000000000 +0100
@@ -80,7 +80,7 @@
void
__gnat_expect_portable_execvp (int *pid, char *cmd, char *argv[])
{
- *pid = (int) spawnve (_P_NOWAIT, cmd, argv, NULL);
+ *pid = _spawnvp (_P_NOWAIT, cmd, argv);
}
int
@@ -168,8 +168,7 @@
__gnat_expect_portable_execvp (int *pid, char *cmd, char *argv[])
{
*pid = (int) getpid ();
- /* Since cmd is fully qualified, it is incorrect to call execvp */
- execv (cmd, argv);
+ execvp (cmd, argv);
_exit (1);
}
@@ -308,8 +307,7 @@
__gnat_expect_portable_execvp (int *pid, char *cmd, char *argv[])
{
*pid = (int) getpid ();
- /* Since cmd is fully qualified, it is incorrect to call execvp */
- execv (cmd, argv);
+ execvp (cmd, argv);
_exit (1);
}
The comment that it is incorrect to call execvp is in fact incorrect itself. The
specification of execvp allows to call it with a fully qualified pathname and in
fact does exactly what is needed here.
With all my suggested patches applied, the handling of spawning child processes
and the way the command itself is passed to the child process are made
consistent.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20075
