Please let us know if you run into any problems displaying pages.
I'm not suggesting you change it back, since it obviously helps most people, but anyone wanting to use this in an environment of fast connections might want to do some testing first.
Jonathan, The version of rlreturnz I have has the IE workaround + some changes I made (handling i18n charset stuff). You can download the code we are using at http://xarg.net/download/rlreturnz.tar.gz
I don't remeber exactly where I got the original code from, I think I got the link from a thread here though.
If anyone else has fixes they should post them.
Maybe mozilla and safari on OS X don't deal as well with gzipping, or maybe it's just regular OS X speed
And if you haven't tried out OS X recently, you might be surprised how much better it performs than earlier versions. I'm quite happy with it as my developmental computer.
The crucial lines from the mod_gzip.c section of a httpd.conf file are:
mod_gzip_item_exclude rspheader WWW-Authenticate:.*
mod_gzip_item_exclude reqheader "User-Agent: Mozilla/4\.0[^ ]"
I was looking into that again today since this thread started up. I'll do it tomorrow (today actually) given ok from the core team.
Sorry for the delay, it is all ready to go.
The second is, of the 100k hits in the openacs log only 55 actually trigger the old mozilla rule and of those I would say probably 25 or so are fake UA strings (typically robots probing for formmail.pl holes or other spidery things and if they advertise accepting gzip then thats their problem).
So for the 0.03% of browsers we might cause a problem for and given what those users web experience must be like already, I certainly am not going to go out of my way to fix it.
If you do want to fix it then I would say that you should probably match on "Mozilla/4.0[0-9]" since I also see things like Mozilla/4.0+(compatible;+MSIE+6.0;) and Mozilla/4.0+(compatible;) (spiders I would assume).
I dug out the link to the explanation of the authentication problem, not sure if it applies to AOLserver.
Also mod_gzip, currently hosted at SourceForge, includes some additional details of the broken browsers in the included HTML docs. You also want to exclude stand alone js and css files:
mod_gzip_item_exclude file \.js$
mod_gzip_item_exclude file \.css$
You've got a point there re. not taking NS4.0* into account in the general case. Also the current mod_gzip docs do suggest a more strict regexp for matching the way you describe:
mod_gzip_item_exclude reqheader "User-agent: Mozilla/4.0"
The reason why I am slightly concerned about this, is that I am working on a project, where the target audience statistically is a low income demographic. I know for a fact that many of the persons, who might be interested in what I am working on, still do use very old machinery on modem connections. Ie. Windows 95 is still in use out there with who knows what for a browser. So gzipped content would be especially valuable for the clients on modems in those cases where it works reliably.
On the .js, .css issue in openacs now we don't really ns_return any .js or .css so it's sort of a nonissue (although I can see doing dynamic css at least). I am not sure if the .js and .css problem is addressed by the IE padding (I think it is, certainly the printing problems are).
Same thing and WWW-Authenticate, most openacs sites don't really use it (and for those that do I expect it is during devlopement and most of the people looking have modern browsers which just do the right thing).
Maybe someone will address those issues, but so far they don't really seem to have been a problem (and rubylane for example gets a lot of traffic and I would think it probably sees more old browsers than most sites and Jim Wilcoxson has said they don't have problems with the gzip encoding).
Is that the version I should use, or should I use Jeff's version? I understand both are based on Jim's version, but does it matter which one I use?