[9fans] rc environment variable flush

[9fans] rc environment variable flush

[cinap_lenrek]

x=foo; @{x=bar}; echo `{cat /env/x}

yields "foo" but

x=foo; echo; @{x=bar}; echo `{cat /env/x}

yields "bar"

the problem is that the environment variables are
only flushed to /env on exec, but not on fork.

in the first example, after executing the
subshell, x is still marked as "modified" by
us and the cat/echo will cause the old value
to be flushed and the changed value from the
subshell be overridden.

havntfork.c does flush the variables on fork tho.

i think it would be more consistent todo this for
havefork.c as well.

is there a good reason not to?

the following patch seems to work fine:

diff -r 5a4a76cd3657 sys/src/cmd/rc/havefork.c
--- a/sys/src/cmd/rc/havefork.c	Mon Aug 19 16:07:07 2013 +0200
+++ b/sys/src/cmd/rc/havefork.c	Mon Aug 19 21:31:06 2013 +0200
@@ -12,10 +12,12 @@
 	int null = open("/dev/null", 0);
 	int pid;
 	char npid[10];
+
 	if(null<0){
 		Xerror("Can't open /dev/null\n");
 		return;
 	}
+	Updenv();
 	switch(pid = rfork(RFFDG|RFPROC|RFNOTEG)){
 	case -1:
 		close(null);
@@ -45,10 +47,12 @@
 	int lfd = p->code[pc++].i;
 	int rfd = p->code[pc++].i;
 	int pfd[2];
+
 	if(pipe(pfd)<0){
 		Xerror("can't get pipe");
 		return;
 	}
+	Updenv();
 	switch(forkid = fork()){
 	case -1:
 		Xerror("try again");
@@ -91,6 +95,7 @@
 		Xerror("can't make pipe");
 		return;
 	}
+	Updenv();
 	switch(pid = fork()){
 	case -1:
 		Xerror("try again");
@@ -152,6 +157,7 @@
 	char name[40];
 	int pfd[2];
 	int sidefd, mainfd;
+
 	if(pipe(pfd)<0){
 		Xerror("can't get pipe");
 		return;
@@ -164,6 +170,7 @@
 		sidefd = pfd[PRD];
 		mainfd = pfd[PWR];
 	}
+	Updenv();
 	switch(pid = fork()){
 	case -1:
 		Xerror("try again");
@@ -191,6 +198,8 @@
 Xsubshell(void)
 {
 	int pid;
+
+	Updenv();
 	switch(pid = fork()){
 	case -1:
 		Xerror("try again");

--
cinap

Re: [9fans] rc environment variable flush

[erik quanstrom]

On Mon Aug 19 15:37:24 EDT 2013, cinap_lenrek@gmx.de wrote:
> x=foo; @{x=bar}; echo `{cat /env/x}
>
> yields "foo" but
>
> x=foo; echo; @{x=bar}; echo `{cat /env/x}
>
> yields "bar"
>
> the problem is that the environment variables are
> only flushed to /env on exec, but not on fork.
>
> in the first example, after executing the
> subshell, x is still marked as "modified" by
> us and the cat/echo will cause the old value
> to be flushed and the changed value from the
> subshell be overridden.
>
> havntfork.c does flush the variables on fork tho.
>
> i think it would be more consistent todo this for
> havefork.c as well.
>
> is there a good reason not to?

here are three reasons.  i am not sure if any of them are
good
- it is going to be nearly impossible to
tell if this behavior is depended on somewhere.
- it also potentially has performance implications.
- subshells can also run asynchronous, asynchronously
modifying the environment.  don't we still have a race?

- erik

Re: [9fans] rc environment variable flush

[cinap_lenrek]

on races... normal forks will all share the /env environment but
not the in memory variables of rc. so when we would normally fork
whoever does an exec (flush) first will override what the values of the
/env variables are, *independent* of the variables that where
actually modified *in* the process.

when we flush *before* fork, then at least both processes start out
with marked clean in memory variables and the processes will flush
only the things they actually change.

--
cinap