On Wed, 2007-05-02 at 11:18 +0200, Alexander Todorov wrote: > Jeremy Katz wrote: > > On Fri, 2007-04-20 at 13:34 +0200, Alexander Todorov wrote: > >> * anaconda.process_and_launch_dogtail_script.diff > >> The biggest patch of all. This adds a --dogtail option to the command > >> line options for anaconda. Checks all possible conditions specifying a > >> dogtail script: kickstart, boot cmd line, anaconda cmd line option. > >> Downloads the script, fork and execute it before anaconda.intf.run(). > > > > The boot command line should just lead to the option being passed on the > > anaconda command line from the loader. See loader2/loader.c's handling > > of the various args like resolution, xdriver, etc. > > So there is no need to check both? Only one of them is ok, right? > Do I have to patch loader.c to parse dogtail=url and pass it to anaconda > script? I suppose I will have to. Correct to all of the above. > > We'll also eventually need to have the filename not be > > hard-coded as that opens things up to race conditions, especially as > > anaconda starts being run on "live" systems in cases like the live CD > > install. > > > Done, now uses tempfile.mkstemp to store the tescase in > /tmp/ramfs/testcase.py.XXXXXX Sounds good. /me makes a mental note to blast /tmp/ramfs from orbit during F8 since everything in /tmp is really a ramfs now... > >> * gui.py.always_load_a11y.diff > >> Removed checking for dogtail. Always loads accessibility. > > > > What's the reasoning for always loading? Loading when needed instead is > > a lot nicer as it helps to reduce the memory footprint in cases where > > it's not needed. > > > in /usr/bin/anaconda > line 863: anaconda.setInstallInterface(opts.display_mode) > Imports gui.py and gtk if graphical install is selected. > > line 926: instClass.setInstallData(anaconda) > Parses kickstart data and sets instClass.ksdata.dogtail. > > I was not sure how to modify anaconda to parse ks before loading gui > without breaking something. That's why accessibility is loaded by default. > > In gui.py only boot cmd line is checked. If dogtail command is added in > kickstart it is not checked. > > What is the prefered way to solve this issue? Should I parse kickstart > file and check only for dogtail command just before `import gui' or > leave it like that? Or should I make more modifications and have ks > parsed only once before `import gui'? Hmmm, I guess I hadn't considered the case of just setting the dogtail info in the kickstart config. I'll need to think on if that makes sense. And also probably just see what the overhead is for loading the a11y modules. One thing that could work would be to notice the dogtail directive in the loader and then be able to pass it as an argument rather than parsing it in the second stage. That has other trade-offs, though. Chris -- any opinions? > >> * scripts.upd-instroot.dogtail_files.diff > >> Added code to extract the missing files into stage2.img. > >> Works fine, already sent to anaconda-devel-list. > > > > Why are the gconf bits needed here? At least in the past, loading the > > modules was good enough to get things initialized and that was all that > > the gconf key affected. ie, what's changed? > > > > In dogtail/utils.py [snip] > It may be easier to have a patch against dogtail that returns always > true for isA11yEnabled() and does not use GConf Ugh, yeah. dogtail really shouldn't be checking for a11y like this. It'll break under a number of circumstances. Jeremy
