Samuel Thibault <[EMAIL PROTECTED]> writes: > Hi, > > Mario Lang, le Fri 16 Feb 2007 14:23:31 +0100, a écrit : >> Willy, are you still willing to help the speakup project to do the final >> cleaning up so that mainline submission could happen? > > I wouldn't say "final": the way speakup access serial ports is to be > completely rethought. The current way it is done will just have a lot > of chances to crash if you access the serial device from a userspace > program, for instance.
I am aware of these problems, and that is exactly why I was never in favour for a speakup-patched kernel by default. The way speakup currently is implemented is IMO not suitable for the average server system, it is too hackish. I've also observed spontaneous slowdowns of my computer when I was using speakup with a serial-connected hardware speech synthesizer. Always when the synth was supposed to talk a lot, so speakup is definitely not the right thing for people with latency requirements (as is the case for me, since I do work with audio equipment under Linux). Thats also why I finally dropped maintainance, since speakup was coming in the way of my own work. The upstream authors were not believing my symptom description, so things fell silently asleep. > I doubt any kernel hacker would accept that mainline. That is why willy originally volunteered to help the speakup project, to find the things that need to be addressed, and show/help the speakup team to address them. OTOH, the way serial IO is currently done in speakup is on the one hand very ugly and hackish, OTOH does it also provide a "feature". If the kernel crashes but speakup somehow can still run, the user can review the screen and read the error message. If speakup would use some kernel internals to access the serial port (instead of the direct outp calls now), the chance is higher it would be taken down by a kernel crash as well. I guess that was one motivation for the maintainers to actually implement it like this, to keep it as much as possible independent from the actual running kernel... > (That would also permit to use USB-to-serial adapters) Yes, that is also a long-standing problem. Users of newish laptop have complained about this since years. >> Back when I was still maintaining a speakup enabled kernel >> in Debian, I bought a hardware speech synthesizer especially >> for testing speakup support for it. > > Mmm, I'd say the best way would rather to write a "fake" speakup driver, > that outputs plain text on the serial port. That would let _everybody_ > test speakup (including all kernel hackers), not only people that have > the hardware. I am not sure this is going to be helpful. If you want to test a talking application, you'll have to have it talk, otherwise you can't test it really. BTW, for those that dont have a hardware speech synthesizer, they can still use the speech-dispatcher adaptor provided by speakup. They just dont get very early boot output, thats basically all. -- CYa, Mario | Debian Developer <URL:http://debian.org/> .''`. | Get my public key via finger [EMAIL PROTECTED] : :' : | 1024D/7FC1A0854909BCCDBE6C102DDFFC022A6B113E44 `. `' `- <URL:http://delysid.org/> <URL:http://www.staff.tugraz.at/mlang/>
