On 12/8/10 3:40 PM, Andrew Garrett wrote:
On Wed, Dec 8, 2010 at 3:11 AM, Trevor
Parscal<tparscal(a)wikimedia.org>wrote;wrote:
This particular change (the "don't
delete line breaks" part of r73196)
should be reverted, Tim's good changes should be pushed upstream, and we
should be using a standard JSMin distribution whenever possible.
I think the point is that it makes bug reports that much easier to
understand when they're reported by somebody who isn't in debug mode. This
has to be traded off against the performance advantages of removing blank
lines. Is there any data on how significant the performance improvements
are?
We already had this debate, in epic fashion, around September 30 - Oct
1st 2010.
Rough estimates were traded, that suggested that these hacks to JSMin would
- increase the total Javascript weight by 2KB;
- delay the page load by 0.3 seconds for a low-bandwidth user (assume
56K);
- increase the total bandwidth served by Wikimedia by ~600GB or even
over 1TB
(however, at our current scale the bandwidth cost is just $25 to
$50/month).
http://lists.wikimedia.org/pipermail/wikitech-l/2010-October/049760.html
As for what this means I think it depends where you sit. It seems that
cost is not a factor. So we have usability versus simplicity of development.
I normally would say that a website should throw everything they've got
into better user experience. A third of a second is already more than
the difference between a page feeling snappy or sluggish.[1] Or, looked
at the other way, 2KB more of Javascript could add several more features.
If you look at our peers in the Alexa rankings, all of them have used
minification -- sometimes really aggressive minification that results in
Javascript that looks like line noise -- for more than half a decade
now. While it's true we have fewer paid developers than they do, even
piddly little single-developer web sites now do this sort of thing
routinely.
So this isn't something Trevor invented just to vex Wikitech-l with. I
understand his exasperation here that we have to rehash debates that the
rest of the web seems to have settled in 2004 or so.
But then again, we really are different so I'm willing to suspend
judgment here. We're not like Google where maybe only 10 people in the
world fully understand the tricks and compromises involved in making the
front page go (or if we are, we shouldn't be). For us, there's a much
more gradual continuum from inside developer to random tech-illiterate
user. So I'm more willing to trust Tim's judgment that better bug
reports and casual inspection of source code are more important for us.
If not, we can revisit this decision at a later time, perhaps when our
community is more used to the whole Resource Loader concept. We'll
probably get lots of bug reports complaining about the useless newlines,
as we just did. ;)
[1] Of course, Javascript loading trails initial page load, but the
point still stands -- 2KB is a LOT.
--
Neil Kandalgaonkar ) <neilk(a)wikimedia.org>