Re: Web Development with Guile in HB

[Gary Houston]

> From: Alejandro Forero Cuervo <address@hidden>
> Date: Sat, 11 Nov 2000 20:55:50 -0500

> 1. Is there any workaround to having to call a function that never
>    returns (such as gh_enter or scm_boot_guile)?  I know this makes it
>    easier to write the garbage collector but it makes it impossible
>    (impossible, not just very hard and complex) to build support of
>    Scheme as a module/plugin (at least using HB's definition of
>    modules).  Or are the only workarounds stupid hacks such as forking
>    and dedicating one process entirely to execute Scheme code?

The CVS version of Guile has this in NEWS:

** New function: scm_init_guile ()

In contrast to scm_boot_guile, scm_init_guile will return normally
after initializing Guile.  It is not available on all systems, tho.

> 2. How can I make functions such as `display' call my callback
>    functions rather than print to stdout?  In an application server,
>    such as HB, printing to stdout is stupid.  I want to capture
>    whatever output the Scheme code produces and add it to some
>    buffers.  Redefining those functions (how?)?  Or can I call some
>    functions to set my callback function so it gets called rather than
>    something that calls write/printf?

String ports could help with this problem:

from Scheme:
guile> (with-output-to-string (lambda () (display "eionwe")))
"eionwe"

from C:
scm_mkstrport creates a string port (strports.c)
scm_set_current_output_port can set the current Scheme output port to the
string port (ports.c)
scm_strport_to_string retrieves the string port contents (strports.c).

> 3. How can I define procedures (as with gh_new_procedure) in a way
>    such that my callback functions get passed a given void pointer I
>    specify?  This is necessary because HB uses multiple threads to
>    attend requests simultaneously so my functions need to know which
>    was the HTTP request that caused them to be called.  For example,
>    in the previous question, my function needs to know what buffer it
>    should work with.

Since procedures defined by gh_new_procedure only take Scheme values,
you couldn't use them directly as callbacks that take a void *.  You'd
need to install a C callback procedure that converts the void * to a
Scheme value and then somehow finds the right Scheme callback to call.

> 4. Is Guile reentrant/thread safe?   Or should I wrap all calls to
>    Guile functions with a mutex/lock?  In case it can be compiled both
>    ways, how can I find out how it was compiled?

I'm not sure what the current status is, perhaps Guile can only be
used from one thread.  This could be a big problem for what you are
trying to do.

> 5. How can I use multiple environments/contexts/name-spaces?  So if HB
>    executes some code in some file with a few (define)s, only that
>    code can see them.  I guess this is not necessary when parsing
>    input from just one user, but when we have an application server
>    executing code from different files and different users, we need
>    separate contexts for each.  I read some examples in a tutorial but
>    it was extremely outdated.  Could someone point me to an example on
>    how to do it in the latest version?

Probably tied up with the solution to 4, you want each thread to have
it's own module.