fhandlers codebase, magic dirs, etc.

Ronald Landheer info@rlsystems.net
Sun Sep 30 04:56:00 GMT 2001

Hello Robert,

>>>> As I'm going to be working on the opendir(), readdir() and stat()
>>>> problems with magic dirs, I just want to make sure I got some
>>>> things straight, and make some things clear, so here goes:
>>>> First off: as I understand it, the readdir() and opendir() code is
>>>> not part of the fhandlers, but should be, so should be migrated
>>>> there, and the existing opendir() and readdir() implementation
>>>> should call the ones in the fhandlers to take care of the dirty
>>>> work, correct?
>>> check.
>> Oh in any case, I will :) I was just hoping that someone would know
>> very sure and fill in details..
> I meant this in the context of a checklist - "check" or "fail". You
> had it right.
Ah, OK :)

>>>> Depending on the type of fhandler that finds stuff in the requested
>>>> opendir() call, the opendir() wrapper should or should not continue
>>>> to the other fhandler classes' opendir() implementations - e.g. in
>>>> /registry, only registry items will be listed, and no files,
>>>> correct?
>>> the opendir wrapper will look very similar to the open wrapper. That
>>> is to say that there is one and only one fhandler that matches a
>>> specific path. So there is no "continue to" logic at all.
>> That would depend on how they're implemented: if an fhandler handles
>> (and shows) a type of file, and that type of file can occur in
>> different types of directories (say: a real file can exist in /dev)
>> that the "continue to" logic is there. If the fhandler handles a type
>> of directory, and shows everything in it, then there's no "continue
>> to" logic to be followed.
>> I don't know what the best approach is yet - but I guess the second
>> one would seem most obvious..
> I disagree. If a "real" file can exist in /dev, that is the
> responsibility of the dev fhandler to manage that. The dev fhandler
> will still do all the grunt work. A directory is just a file that
> points to other files. So the logic for what fhandler looks after a
> fiven directory is _identical_ as for looking after files.
Then you *do* agree - with the second option: if a fhandler handles a 
type of directory - say, the /dev dir - the fhandler for it shows 
everything in it, and there's no "continue to" logic to be followed.

I hadn't decided between the two yet, but with a bit of thought - i.e. 
the concept of mounting the fhandler for /dev at /dev settling in my 
synapses - I agree with not having the "continue to" logic: it's just a 
matter of who owns the directory, and asking him (the proper fhandler) 
for the data.

>>>> Codebase: I haven't gotten familiar with it yet - I'll be starting
>>>> that now. As a result, the above suggestions may be complete
>>>> bullshit and should be regarded with that possibility in mind. This
>>>> is just what a tired mind got from a discussion on the cygwin
>>>> mailing list..
>>> fairly close. I have one thing to add and that is that the readdir
>>> wrapper needs to add (and when necessary override) the fhandler
>>> returned entries for mounted fhandlers below the open node. An
>>> example: in / we mount a registry fhandler at /registry, and a devfs
>>> handler at /devfs. Opendir (/) goes to the fhandler_base (which IMO
>>> should be split into fhandler_win32 and a real fhandler_base that
>>> has almost no functionality - different discussion though ;}).
>>> lets say that there is a dir called registry in the cygwin root, and
>>> not one called dev. We want ls / to show both registry and dev.
>>> Now, special case 1: readdir (/ (fhandler_base)) returns a
>>> "registry" entry. The readdir wrapper sees that "registry" conflicts
>>> with the mount table entry /registry and replaces any relevant
>>> metadata in the response to the calling app. Such metadata could be
>>> obtained by a call to stat (/registry (which ends up going to
>>> fhandler_registry))
>>> Special case 2: (BTW: I'm assuming readdir() returns unsorted
>>> entries here). readdir (/ (fhandler_base)) returns end-of-list, and
>>> the readdir wrapper notices that "dev" has not been listed. So the
>>> wrapper adds "dev" to the returned list.
>> Here, I would propose a "continue to" strategy - which leaves less
>> real work for the wrapper: just ask each of the classes whether they
>> have anything to add to the list: give them all the list - which
>> starts empty - and have them add to the end. Look for duplicates
>> after the last one et voilà.
> Again, I disagree. Remember that a class is simply code - no real
> data. By asking all the classes you are making this entirely special
> case, and not data driven. You are also making the fhandlers more
> complex internally. By using the mount table, the system becomes data
> driven, not code driven, and less code is needed in each fhandler.
You're right - I don't agree with myself either :)

> * Huh? Not sure of the meaning of these two paragraphs. Time for me to
> read susv2 on this I think.
Never mind - my mind was skipping over a rope while it should be 
balancing a wire.. The "data driven" approach is better - doesn't 
require as much special coding, and wouldn't require the "completeness 
over optimisation" approach to coding. (Though that is generally a 
reasonable approach to start out with).
>>>> Documentation: I have the tendency to document my source code
>>>> thouroughly, which generally provides enough technical
>>>> documentation to build a how-this-works document on, if needed,
>>>> I'll be happy to make such documentation too..
>>>> All that being said, I just have one question left: does it make
>>>> sense for me to checkout the CVS sources of the fhandler_*.cc,
>>>> dir.cc, syscalls.cc, etc. files, or are they not likely to have
>>>> been changed since 1.3.3-2?
>>> they change all the time. The best bet is to CVS checkout the lot.
>>> And then either subscribe to cygwin-cvs and update as needed, or use
>>> cvs diff a fair bit, or just get in the routine of doing cvs update
>>> -Pd before you start hacking each evening/morning/whenever. Update
>>> regularly so that incremental changes don't cause you merge
>>> nightmare.
>>> Also a good idea if chris is willing, is for you to create a cvs
>>> branch for this project, so you can be developing under change
>>> control, which will give you a bit more flexability. Then as things
>>> become stable, you can produce specific patches for specific
>>> achievements/milestones.
>> I'm already subscribed to the cygwin-cvs list - have been for about a
>> week now. I'll be downloading the CVS then..
>> As for making a branch in the repository and allowing me access - it
>> might be a good idea, but I don't know how necessary it will be: I
>> have my own repository here, in which I'll be putting whatever I do
>> for cygwin under version control as well, so making big patches
>> should not be a problem. A branch in the repository is only
>> interesting if others will be working on the same thing (i.e. fixing
>> the magic dirs problems), which is not current (I think?).
> I have some interest in this, and was planning on the occasional peek
> :].
>> I've taken a look at the current stat() implementation. My guess is
>> it can't stay: should be replaced by a call to the stat()s in the
>> fhandlers asking "is this yours?", again, followed by some conflict
>> resolution..
> NO!. Or rather "yes th current stat is win32 only and it has to go,
> but this continuation meme you have will make your life very very
> hard".
With some more second though, I agree this would make my life harder. Though the current stat() implementation should go, replaced by something more general/generic.
I agree it should not be a "does this belong to you?" approach, but I 
guess it should be done by the fhandler handling whatever 
directory/mount point is being looked in..

> Examine _open (). Examine it a little further with my mount table
> alterations. Stat should look identical. There is _no_ continuation
> issue or strategy involved in this architecture. _ever_.
I will - I'll have a look. It's likely to be a lot like what stat() 
should offer..

> I'm happy to draw up a how-fhandlers-work document if needed. (I think
> I grok them, Chris or Corinna will probably want to confirm what I
> write :})
Please do - you probably know the layout already, which means that I 
don't have to start at nothing figuring out how it works.



More information about the Cygwin-patches mailing list