On Mon, Sep 27, 2004 at 03:34:06PM -0700, Matthew Dharm wrote:
> On Mon, Sep 27, 2004 at 11:01:32AM -0700, Phil Dibowitz wrote:
> > On Mon, Sep 27, 2004 at 09:34:04AM -0700, Matthew Dharm wrote:
> > > Lots of work?  Tragic.  But that's what you get when you're the
> > > unusual_devs.h guy.  Handle it as you see best, but the goal is to make
> > > unusual_devs.h as small as possible.
> > 
> > Woah there buddy! I'm okay with the work... I don't recall saying "oh,
> > that's gonna take too much work" - I was just discussing the
> > logistics.
> 
> Sorry.  I'm just a little behind in life.  This past weekend was my first
> break in 26 consecutive days of work.  I guess I'm getting a little bit
> punchy.

No worries. I do that a lot. =) Me after 2 or 3 consecutive 80-hour
work weeks is not fun to be around.

> I'm hesitant to "minimize" this much.
> 
> I want a value that will help many of the FIX_INQUIRY devices we have.  I
> want to cut down on people having to add entries to unusual_devs.h in the
> general case.  I want to improve the experience of most people (i.e. "my
> device Just Works(tm) out of the box with Linux").

Absolutely. Let me elaborate a little more, because I don't think I
was very clear.

First and foremost I support the idea of the delay. The Just Works is
something I see as important to Linux. But I also don't accept
anything without weight the consequences. You've been doing this a lot
longer than I have, and I respect your opinion. I'm simply discussing
here.

As far as the delay, here's my opininon:

1. If 100ms delay takes care of 99% of devices and a 5 second delay
takes care of 99.1% of devices, I _strongly_ am in support for the
100ms

2. If 5s handles 70% of devices and 4 seconds handles 30% of devices,
we need 5 second -- and I'd say lets investigate 6 or 7 seconds.

In other words, we want that intersection of smallest delay for
greatest accomplishment.

Devices work a lot quicker than humans - in _most_ cases. I have a
phone that I *have* to specify a sleep(4) in the write code of
multisync in order for it to work. sleep(3) doesn't work. But that's
an *exception* to the rule.

In most cases 100ms is a decent chunk of time to a modern electronic
device. 1 second is usually a LOT of time. Whenever you start
introducing arbitrary delays in the kernel, IMO it needs to be done
carefully, minimally, and precisely.

> Given that this 5 seconds runs concurrently with other things, I don't see
> it as being a big delay at boot.

Agreed. Not huge. But still not something to just forget about.

> Also, given that people who really want to squeeze for the last few seconds
> out of boot can easily make the less-than-one-line change to alter the
> default delay (or just change it via sysfs from their initrd before they
> load the HCD), I don't see a compelling reason to lower it.

Hmm. Perhaps this would be a good place for a boot param? I've always
viewed initrds as pretty ugly.

> I see the idea here to improve life greatly for many people at a small cost
> to few people (those who care about those few seconds).  Since those few
> people can perfectly easily squeeze those seconds out at either
> compile-time or runtime, there just isn't a downside that I really see.

No. I agree, the delay idea (assuming it works for a large portion of
devices) is a GOOD thing. I'd hate to use a 5 second delay though if 1
second does the same amount of good though. Ya know?

-- 
Phil Dibowitz                             [EMAIL PROTECTED]
Freeware and Technical Pages              Insanity Palace of Metallica
http://www.phildev.net/                   http://www.ipom.com/

"They that can give up essential liberty to obtain a little temporary
safety deserve neither liberty nor safety."
 - Benjamin Franklin, 1759

Attachment: signature.asc
Description: Digital signature

Reply via email to