The "ikwiki.cgi?page=index&do=edit" function has a problem when running with thttpd or mini-httpd: for some reason the headers ikiwiki outputs are transmitted as the page content. Surprisingly, the "do=prefs" function works as expected.

Here is what it looks like in iceweasel:

Set-Cookie: ikiwiki_session_apnkit=99dad8d796bc6c819523649ef25ea447; path=/
Date: Tue, 14 Aug 2007 17:16:32 GMT
Content-Type: text/html; charset=utf-8

<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN"

Ikiwiki runs fine with boa.


It doesn't work for signin either. What is the reason for these "header => 1" in FormBuilder initialisations? Why do they appear two times with conflicting values in the very same hashes?


Clearly those duplicate header settings are a mistake. But in all cases, the header => 0 came second, so it should override the other value and can't be causing this problem. (cgi_signin only sets it to 0, too).

What version of formbuilder are you using? If you run ikiwiki.cgi at the command line, do you actually see duplicate headers? I don't:

joey@kodama:~/html>REQUEST_METHOD=GET QUERY_STRING="page=index&do=edit" ./ikiwiki.cgi
Set-Cookie: ikiwiki_session_joey=41a847ac9c31574c1e8f5c6081c74d12; path=/
Date: Tue, 14 Aug 2007 18:04:06 GMT
Content-Type: text/html; charset=utf-8

<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN"

Do thttpd and mini-httpd perhaps not realize that Set-Cookis is the start of the headers? --Joey

Thanks for your help: I think I found the problem! Ikiwiki outputs (in my case) the following error message on stderr, followed by an empty line:

/srv/ikiwiki/wc/index.mdwn:  (Not a versioned resource)

Probably thttpd and mini-httpd read stderr as well as stdout, while apache and boa don't. When using a shell-script wrapper as the CGI, which redirects ikiwiki's error output to /dev/null, it works better.

The edit still fails to commit, because in my wiki, index.mdwn is pulled from the base wiki and somehow ikiwiki wants to change it rather that create it.


If thttpd and mini-httpd interpret CGI's stderr as stdout, then they're not properly following the CGI spec, and will break with tons of cgi scripts besides ikiwiki. And of course there are many many cases where ikiwiki might output to stderr, and that's the right thing to do. So I don't see any way to address this in ikiwiki. --Joey

(reported as Debian bug #437927 and Debian bug #437932) --JeremieKoenig

Marking done since it's not really an ikiwiki bug. --Joey

I'm using boa and getting some odd behaviour if I don't set the umask option in the config file. Editing a page through the web interface and hitting "Save Page" regenerates the index.html file with no world-read permissions. As a result, the server serves a "403 - Forbidden" error page instead of the page I was expecting to return to.

There are only two ways I found to work around this: adding a umask 022 option to the config file, or re-compiling the wiki from the command line using ikiwiki --setup. Setting up a git back-end and re-running ikiwiki --setup from inside a hook had no effect; it needed to be at the terminal. --Paul

Since others seem to have gotten ikiwiki working with boa, I'm guessing that this is not a generic problem with boa, but that your boa was started from a shell that had an unusual umask and inherited that. --Joey

That's right; once I'd worked out what was wrong, it was clear that any webserver should have been refusing to serve the page. I agree about the inherited umask; I hadn't expected that. Even if it's unusual, though, it probably won't be uncommon - this was a stock Ubuntu 9.04 install. --Paul

(I'm new to wiki etiquette - would it be more polite to leave these details on the wiki, or to remove them and only leave a short summary? Thanks. --Paul)

Well, I just try to keep things understandable and clear, whether than means deleting bad old data or not. That said, this page is a bug report, that was already closed. It's generally better to open a new bug report rather than edit an old closed one. --Joey

Thanks for the feedback, I've tidied up my comment accordingly. I see your point about the bug; sorry for cluttering the page up. I doubt it's worth opening a new page at this stage, but will do so if there's a next time. The solution seems worth leaving, though, in case anyone else in my situation picks it up. --Paul