A former colleague wanted to be able to have an arbitrary action taken
if a job takes too long to complete, which is almost what $REPORTTIME
gives us, but we can't have arbitrary actions taken.
So it seems like what's needed is a hook function (and array).
Below is a patch which I will *not* commit, because it's a first pass to
get feedback. For one thing, it hooks into printtime so is also invoked
when the time pre-command modifier prints output. This could be useful,
but might not be desired; I figured it would be best to pass the
printtime() `desc` parameter as the first parameter to the new hook, but
while that lets folks analyse and short-circuit I think it's putting the
onus in the wrong place.
I'm tempted to make this $REPORTTIME only, by amending printtime() to
take a new bool parameter and renaming the hook to reporttime().
Another issue is that reporttime() (shell-level function) becomes a new
magic function additionally called, and we don't have namespaces for
functions, but I think our normal approach is to accept that it _might_
conflict with scripts, right?
Then there's the parameters needed; I've opted to pass in the desc
(which for $REPORTTIME is the command invoked) in the first parameter
then elapsed/user/system/total time as subsequent parameters. I half
think it might be worth putting a bunch of the various variables used
for $TIMEFMT into a hash, but if I do that I'm likely to want to add an
existshookfunc() check to reduce the work, or create a magic hash which
extracts values and make $TIMEFMT reduce to referencing that hash, or a
new print flag similar to -P which can be used to get the same
expansions as $TIMEFMT.
(Patch also fixes an unused variable complaint)
What do people think? Is this generally useful enough, and should I
amend printtime? How best to pass the various values or make them
available?
I *think* I favour:
1: adding a new parameter to printtime();
2: moving moving the printtime() expansion logic into a function which
can be used by "print -T" or the current $TIMEFMT expansion; this
involves making a new global which can hold the last child_times_t
object around for implicit use;
3: calling the hook "reporttime" and then the hook functions can just
use "print -T" for any data they want.
I can add documentation when things are more firmed, then resend the
patch and go through the zsh/git/listid cycle.
Thoughts?
-Phil
diff --git a/Src/jobs.c b/Src/jobs.c
index 0dbb10b..2ca3bfe 100644
--- a/Src/jobs.c
+++ b/Src/jobs.c
@@ -645,11 +645,13 @@ static void
printtime(struct timeval *real, child_times_t *ti, char *desc)
{
char *s;
+ LinkList funcargs;
double elapsed_time, user_time, system_time;
#ifdef HAVE_GETRUSAGE
double total_time;
#endif
int percent, desclen;
+ char buf[20];
if (!desc)
{
@@ -830,6 +832,19 @@ printtime(struct timeval *real, child_times_t *ti, char *desc)
unqueue_signals();
putc('\n', stderr);
fflush(stderr);
+
+ funcargs = newlinklist();
+ addlinknode(funcargs, "printtime");
+ addlinknode(funcargs, desc);
+ sprintf(buf, "%4.2f", elapsed_time);
+ addlinknode(funcargs, ztrdup(buf));
+ sprintf(buf, "%4.2f", user_time);
+ addlinknode(funcargs, ztrdup(buf));
+ sprintf(buf, "%4.2f", system_time);
+ addlinknode(funcargs, ztrdup(buf));
+ sprintf(buf, "%4.2f", total_time);
+ addlinknode(funcargs, ztrdup(buf));
+ callhookfunc("printtime", funcargs, 1, NULL);
}
/**/
@@ -1714,7 +1729,9 @@ static int hackspace;
void
init_jobs(char **argv, char **envp)
{
+#ifndef HAVE_SETPROCTITLE
char *p, *q;
+#endif
size_t init_bytes = MAXJOBS_ALLOC*sizeof(struct job);
/*
Attachment:
pgpS9NoYNQzia.pgp
Description: PGP signature