david reid wrote:
httpd uses pcre via a wrapper already, so moving it into apr-util makes a degree of sense right there.
Taking a look at httpd's wrapper, my immediate question is: why is httpd re-wrapping pcre's native interface in a posix-style interface itself instead of simply using pcre's own posix-style interface? I do not know enough to say for sure, but perhaps httpd's wrapper should simply be deprecated and go away, rather than be moved into apr-util.
Windows doesn't have builtin regex support, so providing regex via apr-util allows people to write more portable code. A good thing.
I do not understand why "uses pcre via apr-util" should be considered more portable thatn "uses pcre"?
There are places within apr-util where using regex would be useful. Knowing that I can simply write code that uses regex functionality, confident in the knowledge that lovely, useful apr-util provides it is another good thing.
If apr-util's configure and windows-buildsystem were to arrange to link with libpcre, then you could use it without first enclosing it in an apr-ified wrapper. What extra benefit does the apr-ified wrapper bring?
Reducing the dependencies for httpd is a good thing.
I'm totally confused about the above sentence. How does simply moving the libpcre dependency to being via apr-util reduce the dependencies of httpd?
Providing an interface that provides more of the libpcre extended functionality seems like a good thing.
But why not use the pcre interface directly?
Allowing those using apr-util to benefit from having builtin regex support is a good thing. Asking them to include yet another library is a bad thing.
On the other hand, we've already come to the conclusion that an ever-increasing list of dependencies for apr-util is a bad thing, and that apr-util needs to modularize to remain useful. Therefore, we could suppose that 'apr-regex' should be a new, self-contained, library.
Asking people to learn the intracies of libpcre (not as simple as they seem as I discovered) to simply use a regex pattern in their code is a bad thing.
Interesting. If the libpcre API is a pain to work with for simpler applications, then there is a real opportunity for a simpler layer to be useful. My first instinct, though, would be to suggest that such a thing might be better placed as part of the pcre project itself.
Adding libpcre as a required dependency of apr-util without providing some way of including the library within apr-util's build (for pcre challenged systems) is a nasty thing - and one that I'm surprised people would be happy with.
Did anyone say that?
Most of the above seemed to be self evident, so the high level of resistance I saw to my proposal surprised me.
Hopefully I've explained why it isn't self-evidently advantageous to me.
The fact that people were happy to complain on IRC but not post to the list worried me.
I can only speak for myself here, but I'd assumed that having noted to you directly that I thought the idea needed a bit of justification, you'd go ahead and post some to the list. I didn't feel any need to post to the list, since I'd already delivered my request for more info to you via IRC.
Why did I withdraw my patch from discussion? The level of resistance I saw was quite high and given how many things I'm working on and the fact that the regex support isn't high on any list of things to be done, I decided to cut it from my todo list - freeing my time up for other things. I'd rather be doing productive things than debating pro's and cons of something that people don't want.
I don't not want it - rather, I don't understand why it should be wanted, and am interested to discover the reasons for which you want it, which I do not see.
It's clear this isn't going to be the trivial httpd->apr-util relocation job which it looks like you were hoping for, so I guess this thread will fizzle out. Hopefully, I've at least managed to explain why reaction wasn't positive.
Max.
signature.asc
Description: OpenPGP digital signature
