Mercurial > mplayer.hg

<HTML>
<BODY BGCOLOR=white>

<FONT face="Verdana, Arial, Helvetica, sans-serif" size=2>

<P><B><I>In medias res</I></B></P>

<P>There are two major topic which always causes huge dispute and flame on the
<A HREF="http://www.MPlayerHQ.hu/cgi-bin/htsearch">mplayer-users</A>
mailing list. Number one is of course the topic of the</P>

<P><B><I>GCC 2.96 series</I></B></P>

<P><B>Also read <A HREF="gcc-2.96-3.0.html">this</A> text !!!</B></P>

<P>The <I>background</I> : there were/are the GCC <B>2.95</B> series. The
best of them was 2.95.3 . Please note the style of the version numbering.
This is how the GCC team numbers their compilers. The 2.95 series are good.
We never ever saw anything that was miscompiled because of the 2.95's faultiness.</P>

<P>The <I>action</I> : <B>RedHat</B> started to include a GCC version of <B>2.96</B>
with their distributions. Note the version numbering. This should be the GCC
team's versioning. They patched the CVS version of GCC (something between 2.95 and 3.0)
They patched it very deep, and used this version in the distrib because 3.0
wasn't out at time, and they wanted IA64 support ASAP (business reasons).
Oh, and GCC 2.95 miscompiles bash on the s390 architecture (there is
no RedHat distribution for s390..) .</P>

<P>The <I>facts</I> : <B>MPlayer</B>'s compile process needs the
<CODE>--disable-gcc-checking</CODE> to proceed upon detecting a GCC version of
2.96 (apparently it needs this option on <B>egcs</B> too. It's because we don't
test <B>MPlayer</B> on egcs. Pardon us, but we rather develop <B>MPlayer</B>).
If you know <B>MPlayer</B>, you should know that it has great speed. It
achieves this by having overoptimized MMX/SSE/3DNow/etc codes, fastmemcpy, and
lots of other features. <B>MPlayer</B> contained MMX/3DNow instructions in a
syntax that all Linux compilers accept it... except RedHat's GCC (it's more
standard compliant). It simply <B><I>skips</I></B> them. It doesn't give
errors. It doesn't give warnings. <B>And</B>, there is Lame. With gcc 2.96, its quality check
(<CODE>make test</CODE> after compiling) <I>doesn't even run !!!</I>
But hey, it compiles bash on s390 and IA64.</P>

<P>The <I>statements</I> : most developers around the world begun having
bad feelings about RedHat's GCC 2.96 , and told their RedHat users to
compile with other compiler than 2.96 . RedHat users' disappointment slowly
went into anger. What was all good
for, apart from giving headaches to developers, putting oil on anti-RedHat
flame, confusing users? The answer, I do not know.</P>

<P><I>Present age, present time</I> : RedHat says that GCC 2.96-85 and above
is fixed, and works properly. Note the versioning. They should have started
with something like this. What about GCC 2.96.85 ? It doesn't matter now.
I don't search, but I still see bugs with 2.96 . It doesn't matter now,
hopefully now <B>RedHat will forget about 2.96</B> and turn towards <B>3.0</B>.
Towards a deep patched 3.0...
</P>

<P><I>What I don't understand</I> is why are we hated by RedHat users for
putting warning messages, and stay-away documents in <B>MPlayer</B> .
Why are we called "brain damaged", "total asshole", "childish" by
<B>RedHat users</B>, on our mailing list, and even on the <B>redhat-devel</B> .
They even considered forking <B>MPlayer</B> for themselves. RedHat users.
Why? It's RedHat that made the compiler, why do <U>you</U> have to hate us?
Are you <U>that</U> fellow RedHat worshippers? Please stop it. We don't hold
a grudge against users, doesn't matter how loud you advertise its contrary.
Please go flame Linus Torvalds, the DRI developers (oh, now I know why
there were laid off by VA!), the Wine, avifile. Even if we are arrogant,
are we not the same as the previously listed ones? Why do <B>we</B> have
to suffer from your unrightful wrath?</P>

<P>I'm closing this topic. Think over it please. I (Gabucino) personally begun
with <A HREF="http://www.redhat.com">RedHat</A>, then used Mandrake (sorry I
don't know their URL), now I have <A
HREF="http://www.linuxfromscratch.com">LFS</A>. Never held a grudge against
RedHat or RedHat users, and I still don't. Hate is only comfortable. It
won't bring you anywhere.</P>

<P><B><I>Binary distribution of MPlayer</I></B></P>

<P>Tons of users asked us about this. For example Debian users tend to say: Oh,
I can <CODE>apt-get install avifile</CODE>, why should I <B>compile MPlayer</B> ?
While this may sound reasonable, the problem lies a bit deeper than
those-fuckin-MPlayer-developers-hate-gcc-2.96-and-RedHat-and-Debian.</P>

<P>Reasons: <B>Law</B></P>

<P><B>MPlayer</B> describes the <U>sourcecode</U>. It contains several files with incompatible
licenses especially on the redistribution clauses. As source files, they are
allowed to coexist in a same project.</P>

<P>Therefore, <U>NEITHER BINARIES NOR BINARY PACKAGES OF <B>MPlayer</B> ARE ALLOWED TO EXIST SINCE
SUCH OBJECTS BREAK LICENSES</U>. PEOPLE WHO DISTRIBUTE SUCH BINARY PACKAGES ARE
DOING ILLEGAL ACTIVITIES.</P>

<P>So if you know somebody who maintains a binary package then forward her/him
this text and (ask him to) contact us. What (s)he is doing is illegal and IT IS
NO LONGER <B>MPlayer</B>, but <U>his/her</U> mplayer. If it breaks, it is
his/her fault. Don't come and cry on the <B>MPlayer</B> mailing lists, you will
most likely be blacklisted.</P>

<P>Reasons: <B>Technical</B></P>

<P>
<UL>
  <LI><B>MPlayer's</B> speed (MMX, SSE, fastmemcpy, etc) optimizations are
    determined during compilation. Thus a compiled binary contains very
    processor-specific code. An <B>MPlayer</B> binary compiled for K6 will die
    on Pentiums and vice versa. This has to be workarounded by runtime
    detection, which is not an easy thing to do becase it causes massive speed
    decrease. If you don't believe (it was explained in details 10000 times on
    mplayer-users, search the archive), solve it and send us a patch. Someone
    begun work on it, but disappeared since then.</LI>
  <LI><B>MPlayer's</B> video/audio system is not plugin based. It is compiled
    into the binary, thus making the binary depend on various libraries (the
    GUI depends on GTK, DivX4 depends on libdivxdecore, SDL depends on libSDL,
    every SDL release contains an unique bug that has to be workarounded during
    compiletime, X11 output compiles differently for X3 and X4, etc). You may
    say: yes, let's make 30 versions of downloadable binaries! We won't. We
    will make these stuff pluggable in the future.</LI>
</UL>

</HTML>
author	alex
date	Sun, 13 Jan 2002 14:52:58 +0000
parents	fba04e21ddec
children	23abc0e262e3