Re: [Qemu-devel] [5083] PPC: Switch a few instructions to TCG

[Andreas Färber]

Am 26.08.2008 um 00:25 schrieb J. Mayer:

<snip>

The sole reason I replied to your original message was that I did in  
fact perceive it as inappropriate and very rude towards Aurelien.

I did not argue about your code or activity, nor did I ask you to  
resign, I just called on you to be a little more reasonable than this  
don't-touch-my-code attitude and resulting noise. In doing so, I tried  
to show you some valid reasons why someone would commit in "your" code  
in an Open Source project, starting with that you did not write to the  
list what you supposedly did (maybe a language problem).
Aurelien could have posted a patch first, yes, but still that is no  
reason for such an eruption of anger. Instead you could've stated the  
pure facts, mentioning what exactly conflicted, and tried to cooperate  
on fixing it instead of commanding your fellow committer to withdraw  
any changes of his just because they're in "your" code.

And I am not aware of any insult from my side.

Concerning OHW, it has a lot to do with the ppc targets because ppc- 
softmmu and ppc64-softmmu both use it exclusively, for all machine  
variations I tried. QEMU does not ship any other ppc ROM. No one has  
yet reported success in replacing it with any other OpenFirmware  
implementation to my knowledge, Laurent seemed closest recently. If  
someone somewhere has supposedly succeeded replacing OHW somehow, I  
ask you, does knowing that really help users of QEMU?
CD/DVD booting is broken, making it impossible to set up a new ppc VM.  
There is no downloadable test image for ppc to check, so I recently  
tried a Haiku/ppc image and it didn't boot either. IDE status  
mismatches for both, despite the relaxed binary-only checks which your  
commit message does not describe. Earlier, pre-TCG, I reported getting  
to a yaboot prompt but it errored out shortly after with lots of  
apparently OHW related errors. So even if OHW is conceptually a  
different project, then, like for Bochs BIOS, QEMU relies on it  
working and needs a usable version out-of-tree or a diff against the  
out-of-tree version to apply the necessary fixes for QEMU.
In an Open Source world I can't expect you to respond to my personal  
problems but since I wasn't doing anything fancy I consider it a  
general problem (it affected the released versions of Q using QEMU  
0.9.0 and 0.9.1, and Fedora 9's KVM, for instance, and there was/is no  
known workaround). If you and anyone else have been happily running  
ppc-softmmu all along then you have not replied that to my messages  
telling me and others how you're doing it, nor replied whether the  
Linux images I had referenced work for you. I have mainly tried OSX/ 
ppc and Solaris/amd64 hosts (also Linux/i386 once to make sure) and  
Debian Etch (40r0, 40r2), Fedora 9 and OpenSUSE 11.0 .iso files, most  
of which were tested to work on real hardware when written to media.
Anyway, that never was a justification for or against committing  
things. It just proves it ridiculous blaming Aurelien for supposedly  
breaking the target with those two TCG commits. Whether he introduced  
a regression or not, it was broken before.

Concerning the order of fixing things, you are right that converting  
dyngen code 1:1 to TCG would carry over any lingering bugs. On the  
other hand, TCG would allow testing on more platforms, likely leading  
to discovery of such bugs at some point.

Whatever you choose to do, I wish you good luck with your projects and  
a little more placidness.

Salutations,

Andreas