[mp1] subprocess_env and non-mod_perl handlers

[John Wittkoski <jo...@aol.com>]

Greetings all,

So I've been trying to find an answer to this in both the archives and 
in the available mod_perl books, but I have not had any luck, so I was 
hoping someone on the list would know the answer.

Generally: how does subprocess_env in mod_perl 1.x behave with 
non-mod_perl handlers?

It seems from the details I can find that subprocess_env is tied to the 
"live" table of the apache process. However, I am seeing behavior that 
seems to indicate otherwise.

Example 1:

I have mod_perl handlers defined for the TypeHandler and FixupHandler 
phases. If in the TypeHandler code I do this:

$r->subprocess_env("TEST" => "value");

Then in FixupHandler I can "see" TEST using:

$something = $r->subprocess_env("TEST");

Which is what you would expect.


Example 2:

Basically the same except I have a C handler defined for TypeHandler and 
  a mod_perl handler defined for FixupHandler. When the C code does:

ap_table_set(r->subprocess_env, "TEST", "value");

The mod_perl handler for FixupHandler doesn't see it using:

$something = $r->subprocess_env("TEST");


If I do the void subprocess_env trick:

$r->subprocess()

before retrieving $something, then it's populated along with the rest of 
the environment.

However, as mentioned in many of the docs/books, this is expensive and I 
really only need the one variable.

I've also tried walking the subprocess_env table in the perl handler, 
but the value set in the C handler is not there.

Are the C API subprocess_env table and the mod_perl API subprocess_env 
table distict until something (like the void call) merges the tables? Is 
this some sort of scoping issue between C and Perl?


Thanks,

    --John




-- 
Report problems: http://perl.apache.org/bugs/
Mail list info: http://perl.apache.org/maillist/modperl.html
List etiquette: http://perl.apache.org/maillist/email-etiquette.html

Re: [mp1] subprocess_env and non-mod_perl handlers

[John Wittkoski <jo...@aol.com>]

Geoffrey Young wrote on 5/3/04, 8:02 PM:

 > > If I do the void subprocess_env trick:
 > >
 > > $r->subprocess()
 > >
 > > before retrieving $something, then it's populated along with the
 > rest of
 > > the environment.
 >
 > that only affects %ENV, not the ability of $r->subprocess_env to grab
 > something from the subprocess_env table.  %ENV is not automatically
 > populated with the contents of subprocess_env, so if you're really
 > talking
 > about %ENV then your Example 2 + subprocess_env() trick sounds right.


Nope, I'm talking about the table itself (via subprocess_env), which is 
why I am confused.


 >
 > >
 > > However, as mentioned in many of the docs/books, this is expensive
 > and I
 > > really only need the one variable.
 > >
 > > I've also tried walking the subprocess_env table in the perl handler,
 > > but the value set in the C handler is not there.
 >
 > that's really strange.  are you sure that you are not removing it in your
 > application someplace?  try tweaking the test tarball I mentioned bit
 > by bit
 > until it has the relevant logic from your code in it.  I can't tell
 > you the
 > number of bugs I've "fixed" by trying to reproduce the bug, only to
 > find I
 > was the bug :)


Understood. :-) Thanks for the code, I'll keep working on it trying to 
narrow down the problem. At least I know that the behavior I'm seeing is 
definitely wrong and it should work without the void subprocess_env call.


 >
 > >
 > > Are the C API subprocess_env table and the mod_perl API subprocess_env
 > > table distict until something (like the void call) merges the
 > tables? Is
 > > this some sort of scoping issue between C and Perl?
 >
 > it shouldn't be, so long as you're hitting the tables directly with
 > $r->subprocess_env.  %ENV is another matter entirely.


Thanks,

    --John





-- 
Report problems: http://perl.apache.org/bugs/
Mail list info: http://perl.apache.org/maillist/modperl.html
List etiquette: http://perl.apache.org/maillist/email-etiquette.html

Re: [mp1] subprocess_env and non-mod_perl handlers

[Geoffrey Young <ge...@modperlcookbook.org>]

John Wittkoski wrote:
> 
> Geoffrey Young wrote on 5/4/04, 4:18 PM:
> 
>  > not really.  but clearly you have some kind of problem.
> 
> Doh! I definitely do, I think I'm it.
> 
> Remember earlier when you said how in debugging you often find that you 
> are the bug? Well, I am the bug in this case. :-)

:)

> 
> In the process of trying to figure out why I wasn't seeing anything in 
> the subprocess_env table, I decided to take a look at the headers_in 
> table. And there it was.
> 
> Turns out that the C handler I am using (for which I have limited docs) 
> is putting it's data into the headers_in table and not the 
> subprocess_env table. (Insert embarrassed big red face here.) Based on 
> everything I've read on handlers and the types of things being populated 
> (i.e. not headers from the client) the headers_in table didn't seem to 
> be the right place so I wasn't looking there.
> 
> Sorry about that.

no problem at all :)

at least you know a little more about both your app and some apache/mod_perl
internals, so it was positive overall.

--Geoff

-- 
Report problems: http://perl.apache.org/bugs/
Mail list info: http://perl.apache.org/maillist/modperl.html
List etiquette: http://perl.apache.org/maillist/email-etiquette.html

Re: [mp1] subprocess_env and non-mod_perl handlers

[John Wittkoski <jo...@aol.com>]

Geoffrey Young wrote on 5/4/04, 4:18 PM:

 > not really.  but clearly you have some kind of problem.

Doh! I definitely do, I think I'm it.

Remember earlier when you said how in debugging you often find that you 
are the bug? Well, I am the bug in this case. :-)

In the process of trying to figure out why I wasn't seeing anything in 
the subprocess_env table, I decided to take a look at the headers_in 
table. And there it was.

Turns out that the C handler I am using (for which I have limited docs) 
is putting it's data into the headers_in table and not the 
subprocess_env table. (Insert embarrassed big red face here.) Based on 
everything I've read on handlers and the types of things being populated 
(i.e. not headers from the client) the headers_in table didn't seem to 
be the right place so I wasn't looking there.

Sorry about that.

Thanks for the help,


    --John




-- 
Report problems: http://perl.apache.org/bugs/
Mail list info: http://perl.apache.org/maillist/modperl.html
List etiquette: http://perl.apache.org/maillist/email-etiquette.html

Re: [mp1] subprocess_env and non-mod_perl handlers

[Geoffrey Young <ge...@modperlcookbook.org>]

> Geoff,
> So I haven't been able to get very far on the code to test this further, 
> but in the Eagle book I noticed this (section 9.1.4):
> 
> "subprocess_env() is only required if you need to change the environment 
> in a subprocess launched by a different handler or module."

what this section is talking about is the ability of mod_perl to alter the
environment of other handlers, such as mod_php.  in those cases, you are
required to use subprocess_env.

> 
> So what would happen if the C module is setting it's own ENV instead of 
> using ap_table_set? 


ou're thinking too hard :)  there is no ENV in C-land, really.  I mean,
there are ways to get to perl's %ENV via C but that's not what you say
you're doing.  so forget about %ENV for the moment - %ENV gets set
specifically my modules (mod_cgi, mod_perl, mod_php) based on the contents
of the subprocess_env table and has no bearing at all if you're accessing
the table directly.

> Would that explain why I can't see the value in the
> perl module using subprocess_env, but when I call void subprocess_env(),
> the value suddenly appears in %ENV and the subprocess table?

not really.  but clearly you have some kind of problem.  unfortunately I've
never seen this before and I don't see anything in the code to suggest what
your problem might be.  and if I can't reproduce it I can't solve it.

here are a few things to try though.

first, see if changing your calls make any difference.  that is, use
$r->subprocess_env->get() and $r->subprocess_env->set() instead of other
forms to remove the tie magic from the equation and see if that helps at all.

next I would try fiddling with PerlSetupEnv to see if that has some hidden
interaction with your reads.

you seem to know some C, so I'd try creating a C content handler and use
that to just dump the subprocess_env table.  you can use mod_example.c as a
guide.  if you can see the variable in your own content handler but not when
mod_perl is the content handler then I guess that at least verifies
something.  (what, I don't know :)

anyway, if you are able to reproduce the problem and roll it up into the
tarball I posted then I could fix whatever it is in mod_perl core that's
causing you harm.

HTH

--Geoff



-- 
Report problems: http://perl.apache.org/bugs/
Mail list info: http://perl.apache.org/maillist/modperl.html
List etiquette: http://perl.apache.org/maillist/email-etiquette.html

Re: [mp1] subprocess_env and non-mod_perl handlers

[John Wittkoski <jo...@aol.com>]

Geoffrey Young wrote on 5/3/04, 8:02 PM:

 > > Example 2:
 > >
 > > Basically the same except I have a C handler defined for TypeHandler
 > and
 > >   a mod_perl handler defined for FixupHandler. When the C code does:
 > >
 > > ap_table_set(r->subprocess_env, "TEST", "value");
 > >
 > > The mod_perl handler for FixupHandler doesn't see it using:
 > >
 > > $something = $r->subprocess_env("TEST");


 > > However, as mentioned in many of the docs/books, this is expensive
 > and I
 > > really only need the one variable.
 > >
 > > I've also tried walking the subprocess_env table in the perl handler,
 > > but the value set in the C handler is not there.
 >
 > that's really strange.  are you sure that you are not removing it in your
 > application someplace?  try tweaking the test tarball I mentioned bit
 > by bit
 > until it has the relevant logic from your code in it.  I can't tell
 > you the
 > number of bugs I've "fixed" by trying to reproduce the bug, only to
 > find I
 > was the bug :)


Geoff,
So I haven't been able to get very far on the code to test this further, 
but in the Eagle book I noticed this (section 9.1.4):

"subprocess_env() is only required if you need to change the environment 
in a subprocess launched by a different handler or module."

So what would happen if the C module is setting it's own ENV instead of 
using ap_table_set? Would that explain why I can't see the value in the 
perl module using subprocess_env, but when I call void subprocess_env(), 
the value suddenly appears in %ENV and the subprocess table?


    --John






-- 
Report problems: http://perl.apache.org/bugs/
Mail list info: http://perl.apache.org/maillist/modperl.html
List etiquette: http://perl.apache.org/maillist/email-etiquette.html

Re: [mp1] subprocess_env and non-mod_perl handlers

[Geoffrey Young <ge...@modperlcookbook.org>]

> Example 1:
> 
> I have mod_perl handlers defined for the TypeHandler and FixupHandler 
> phases. If in the TypeHandler code I do this:
> 
> $r->subprocess_env("TEST" => "value");
> 
> Then in FixupHandler I can "see" TEST using:
> 
> $something = $r->subprocess_env("TEST");
> 
> Which is what you would expect.

indeed.

> 
> 
> Example 2:
> 
> Basically the same except I have a C handler defined for TypeHandler and 
>   a mod_perl handler defined for FixupHandler. When the C code does:
> 
> ap_table_set(r->subprocess_env, "TEST", "value");
> 
> The mod_perl handler for FixupHandler doesn't see it using:
> 
> $something = $r->subprocess_env("TEST");

that's not right.  however, I can't reproduce this with the attached minimal
test case.  see

  http://perl.apache.org/~geoff/subprocess-env.tar.gz

> 
> 
> If I do the void subprocess_env trick:
> 
> $r->subprocess()
> 
> before retrieving $something, then it's populated along with the rest of 
> the environment.

that only affects %ENV, not the ability of $r->subprocess_env to grab
something from the subprocess_env table.  %ENV is not automatically
populated with the contents of subprocess_env, so if you're really talking
about %ENV then your Example 2 + subprocess_env() trick sounds right.

> 
> However, as mentioned in many of the docs/books, this is expensive and I 
> really only need the one variable.
> 
> I've also tried walking the subprocess_env table in the perl handler, 
> but the value set in the C handler is not there.

that's really strange.  are you sure that you are not removing it in your
application someplace?  try tweaking the test tarball I mentioned bit by bit
until it has the relevant logic from your code in it.  I can't tell you the
number of bugs I've "fixed" by trying to reproduce the bug, only to find I
was the bug :)

> 
> Are the C API subprocess_env table and the mod_perl API subprocess_env 
> table distict until something (like the void call) merges the tables? Is 
> this some sort of scoping issue between C and Perl?

it shouldn't be, so long as you're hitting the tables directly with
$r->subprocess_env.  %ENV is another matter entirely.

--Geoff

-- 
Report problems: http://perl.apache.org/bugs/
Mail list info: http://perl.apache.org/maillist/modperl.html
List etiquette: http://perl.apache.org/maillist/email-etiquette.html