Re: Suggestion for file type detection approach

[Rodney Dawes <dobey free fr>]

On Sht , 2003-12-27 at 12:06, Fabio Gomes wrote:
> Em Sáb, 2003-12-27 às 08:31, Adam Williams escreveu:
> 
> > Nautilus ain't the worlds fastest file broweser but this thread makes it
> > sound like there is some just-horrible-borderline-unusable performance
> > issue.  There isn't.
> 
> There are serious performance issues, because a i286 with 4MB of RAM
> running WinFile.exe on Windows 3.x can do *a lot* more than an AMD Duron
> 950 with 256 MB of RAM running Nautilus. This sucks. A file manager is a
> damn simple and basic application. 

Do a lot more in terms of what? File management? Feature set? How well
does IE 5.5 run on that 286 with 4MB of RAM? Please don't compare apples
and pecans in order to spread FUD about performance. There is really no
constant in your comparison. I am guessing that it's not a real
comparison or benchmark either.

> A basic application, intended to run most of the time to do the most
> basic thins in an OS like launching programs, copying and opening
> documents, should not hog the performance or eat all the memory. 

It doesn't hog the performance or eat all of the memory. It's not
competing with Windows 3.1. It's competing with Mac OS X and Windows XP.
The minimum requirement for XP is, I believe, a Pentium processor and
128 MB of ram, while OS X requires a G3 or better and 128MB of RAM.

32483 dobey      9   0 21836  21m  10m S  0.0  2.4   0:33.74 nautilus

21MB usage is very little in the world of computing today. 64K is no
longer enough for any program to run in. Developers working on software,
on machines that were made in the last year or two, can not spend all
their time making sure that the software runs reliably on a 10 year old
machine. It's not worth the time. Nautilus is plenty fast on all of my
machines, including the Powerbook G3 400 with 192MB of RAM. The only
time that Nautilus *ever* ends up eating a lot of resources on my
system, is when it has to generate thumbnails. And yes, I do have the
thumbnailers for PDF, DOC, etc... enabled, which are really slow aside
from whatever Nautilus does.

> The file manager must be lightweight. Or people will continue to rip
> Nautilus off GNOME. Make a poll and you will see how much people rip
> Nautilus off because it is slow or, in the worst case, don't even use
> GNOME at all. I've brought up this discussion because I *use* Nautilus.
> Many people here that say they are against extension-based MIME type
> recognition admit that they don't use Nautilus at all.

Perhaps you haven't followed the development of Nautilus much in the
last 3 years. I'd suggest you go back and install Nautilus 1.0 and
compare it to Nautilus 2.5. You'll see a very noticeable difference in
size, speed, and simplicity. I am against extension-based MIME type
resolution for the sole purpose of it's supposed to be a representation
of the type of content in the file, not what the file is named. I do use
Nautilus, though not exhaustively. However, I do use gnome-vfs rather
exhaustively. And you seem to be solely complaining about MIME type
detection methods. I don't even know why you are mailing any user lists
at all. gnome-list is not for development discussion. I also don't see
why you need to bother application-level developers that don't hack on
the VFS layer. As far as I can tell, from reading your e-mails, and all
the FUD you've been spreading, your mails should have only really been
sent to gnome-vfs-list.

> I ripped off content-based MIME type detection from my GNOME-VFS and now
> I got decent (acceptable) performance in Nautilus. And all my friends
> that use GNOME are asking me to help them do the same. 

I don't think "rip off" is the term you are looking for, here, or
anywhere, in any of your mails. Nobody is stealing the gnome-vfs or
Nautilus code because they think it is slow. If someone has mad
reasonable effort to make it faster and fix the bugs, and their patches
have been refused for silly reasons, then maybe someone might rip it
off. And, if just making gnome-vfs not use the content sniffing for file
types dramatically sped up your system, you were probably doing
something seriously wrong before. I just ran the test-mime test from the
gnome-vfs source, on 619 files. It took less than half a second. I
wonder what is so wrong with content sniffing, that you think it's the
cause of the performance issues and such that you say you have, and have
been spreading FUD and wasting people's time with.

> Users don't care about content sniffing. They simply want decent
> performance. If content sniffing was possible with decent performance,
> it would be a bonus, but users would still not care about removing it.

You are right. Users don't care about content sniffing. They care that
their files are easily recognizable in the file manager and that they
open with the correct application. Having to go through and rename all
their files, and/or change all the file type preferences to make that
happen, is not a reasonable solution for them. Content sniffing of local
files is not slow by any means. I suggest to compare test-mime to file,
rather than comparing Nautilus to Windows 3.1. It would be much more
productive of you to actually create a reasonable benchmarking
mechanism, and present hard cold data, rather than spreading FUD about
what you think is fast on your system. You need to have as many
constants as possible, and as few variables as possible, to create
really usable data. Foo does blah on this machine, while bar does crap
on another machine, is just useless. By all means, go profile *only* the
mime sniffing stuff, and then profile *only* the extension checking
stuff, and then compare them. You'll see that there isn't much
difference. Please go profile all the things that you think are slow,
and write real comparisons. Then come back and present that data as a
reason to do this or that. If it really is as slow as you say it is,
then it needs to be fixed, not removed. Surprising as it may be, things
do occasionally get fixed and released around here. Please stop sending
complaints and suggestions to remove various pieces of functionality,
because you think they are the problem. It's just FUD. It's not a
security issue. There are no, or very little, performance issues with
the content-sniffing method of MIME type resolution. So, please stop
wasting time without real facts, and please stop trying to spread the
FUD. It's not doing anyone any good.

-- dobey