Web lists-archives.com

Re: New httpd-fastcgi virtual package




Xavier Guimard writes ("Re: New httpd-fastcgi virtual package"):
> Le 19/09/2017 à 21:04, Marco d'Itri a écrit :
> > On Sep 19, Xavier <x.guimard@xxxxxxx> wrote:
> >> So FastCGI application could have dependency on it.
> > How would this work?
> > The packages that could use it would still need to ship a configuration
> > file for every web server since there is no common API like
> > /usr/lib/cgi-bin/ .
> 
> You're right, but a FastCGI server needs a webserver as frontend which 
> can expose it. My idea was to use this virtual package for webserver 
> able to establish the link between a browser and a FastCGI server 
> (Nginx, libapache2-mod-fcgid,...)

If the user needs to configure the webserver specially, by writing
config files, or whatever, in any case, then they will not find it
much help to have some random webserver pulled in by the dependency
system.  Even if they have the webserver already, so apt pulls in the
right module, the user will probably have little difficulty figuring
out which webserver module package to install (if one is needed).

In general, it is best if a virtual package name relates to a
_specific_ interface, provided by a specific name or port number or
whatever.  So for example mail-transport-agent means
/usr/sbin/sendmail, not "anything that can be an MTA".

(And indeed httpd-cgi means "executes scripts in /usr/lib/cgi-bin" and
I assume the httpd-wsgi and httpd-wsgi3 virtual packages are similar.)

Is there some way to provide a uniform interface to fcgi programs ?
If so please try to define one, even if it's just "runs fcgi programs
provided in /usr/lib/fcgi-bin" or something.  (I know fcgi is a
terrible prototocol and may make this hard.)

It would also be worth saying something like "Debian webservers should
violate the FCGI specification by executing fcgi programs with stdout
and stderr connected to somewhere where log messages can be written".

(All the fcgi application glue stuff I tested, including the official
reference implementation in C, was completely happy to be run that way.
Executing an executable with no stdout or stderr is a foolish thing to
do and has malbehaviour in some error cases.)

Ian.