Re: Basic linux network questions (long)
> True, but I think it should be noted that there's only one likely situation
> of serious data loss that seems to be known with APM (that's related to DMA
> driven IDE drives operating at the time of suspend).
Everybody in the world who has a "crashes when tries to resume" problem has
experienced a fsck on true-restart. While uncommon this occasionally loses
something.
Unless folks go to herculean efforts (e.g. not syslogging at all, etc) then
the chance is "musical chairs" that the disk will be being hit when you
suspend, whether it is using DMA directly or not.
> There have been a number of reports of serious problems with ACPI, and the
> code is changing quite rapidly too.
>
> > Reminds me of the quote, "if you left deciding when the project was
> > finished to its lead programmer, no software would ever be shrink wrapped."
>
> I see just the opposite, programs that are obviously a longway from what most
> people would call a release state being released because they fulfil the
> needs of the developer (who incidentally doesn't need manuals so doesn't
> notice that the manuals are no good).
Lead programmer's definition of "done" : asymptotically approaching able to
walk his dog and give his cat kitty treats while refilling his coffee/jolt cola
simultaneously.
Fellow programmer's definition of "throw it on the ftp site" : works for them,
didn't crash all 3 of their boxes.
Manuals written in technese with "install is easy, ./configure, make, make
install, have fun!" as their main content are sadly common.
The core of this is that programmers will not be encouraged to Do The Right
Thing - fix it - unless they are pointedly told that "it's borked" and
most especially, how it's borked, which "it" you have loaded, and that there
is someone out there foot-tapping until they un-bork it or hand it off to
someone who will.
> > > Yes, hope and pray that your hard drive is not one of those devices that
> > > is not quite right...
> >
> > standard IDE is one of the early things to do right, or why bother? But
> > at that point I'm just babbling and no longer offering more than an
> > opinion.
>
> It seems that there are still ongoing issues with IDE and APM...
Until we're sure the memory management is untangled, we can't be sure it's
not related to that too. He's just trying to get a system that will actually
suspend (IMHO). Every day we all take our chances that our box will have a
bad-hair nanosecond, and lose its cookies... he just needs to pick something
that will reduce that likelihood to "not every day" or even "not real often".
We've given him two roads to try and I wish him the luck not to need more luck.
> > I'm also not a user running a journaling filesystem for day-to-day use, but
> > otoh, if one *knows* they will crash a lot, maybe ext3 is a good idea...
>
> If you know that you'll have a "crash" being total system outage (IE it
> suddenly stops and all you can do is press reset) then a journalled file
> system is a good idea. If you suspect that it will be of the slow-memory
> corruption variety then a journalled file system may do more harm than good
> by preventing fsck...
ext3 splits the diff, you can force it to be treated like ext2 if you are
brutal enough.
RAID and other hardware/software layer matters do not protect against the
computer being ordered to be stupid, this is true.
> > > I've read that one of the Unix creators once declared that his only
> > > regret in life was not calling it create() instead of creat().
>
> inline int creat(const char *pathname, mode_t mode)
> {
> return create(pathname, mode);
> }
>
> Put that in one of the system header files and you can rename the function.
> Then hack the ELF header of the libc to make a dlsym(handle, "creat") return
> the handle for create().
>
> Worse hacks than this are in libc already...
Ahh, but they weren't at the time, PCs in the "personal" sense didn't exist
yet except if you had soldered one together yourself, and he had principles.
This wouldn't get *rid* of creat() so it's not worth the hack. However that
is *so* far off-debian I'm not following that sub-thread any further.
* Heather Stern * star@ many places...
Reply to: