2 Feb
2010
2 Feb
'10
10:10 p.m.
On 02/02/2010 09:53 PM, Vincent Massol wrote:
On Feb 2, 2010, at 9:42 PM, Sergiu Dumitriu wrote:
Not all CSS is inherently unstable, just the one we have so far. This is
the main reason why I'm pushing for a clean skin that doesn't inherit
the fragility from Dodo + the fragility from Albatross + the fragility
from Toucan. We're responsible for the degree of overlapiness. So far,
we've made the CSS a jungle. The Colibri cleanup helped here.
What can be done to ensure a side-effect free CSS:
- make sure the selectors are specific enough not to have unpredicted
side effects
- make sure the selectors are generic enough so that they continue to
work when the targeted content is in a slightly different position, and
they allow a new element to reuse these styles
- standardize and document z-indexes
- try to reuse existing styles as much as possible (reusing classname,
accepting inherited styles if they are good enough instead of always
overriding everything just for a little touch of personality)
- reduce hacks
- use !important sparingly
On 01/21/2010 09:04 PM, Vincent Massol wrote:
I'm not focused on comment. What I'd like is that we find ways to
reduce regressions.
Whenever I see some CSS change I cringe since I have no clue what's
going to break. For example JV changes some CSS today to fix an
issue. I couldn't help but wonder in my head is this wasn't going to
break some other place. It's probably due to the fact that I don't
have a visual representation in my head about how the different parts
of our CSS affects different parts of the default XE.
In java code we have found ways to cope with this: - comments -
automated tests - compartimented design (OO design)
I wonder if there are ways to improve our resistance to regression on
the CSS side, and that we could put as best practices. All I know is that I see regularly regressions
when devs touch
the CSS and one reason I believe is because CSS files are not
commented so you don't know why something was done in a given way
and you change it, thus breaking some other part.
I doubt that the lack of comments is the cause of more than 2% of
all the skin regressions that occurred. To bring the problem on a
more familiar ground, imagine that you would have to change some
program made of global variables, and not just that, global structs
with overlapping memory zones. And there are 300 such variables.
What amount of comments would you think would be enough to
guarantee that somebody new to the program won't accidentally break
something? Commenting the 49% is like commenting that one of the
variables of such a global overlapping minefield, called
MAX_NUMBER, should have a positive value.
This is how CSS works. The very nature of selectors make it almost
impossible to know what exactly will be side-affected when applied
to a real HTML. I don't often change CSS but when I do (or anyone
else for that
matter), I'd like some assurance that I'm not breaking something
elsewhere (that breaks through complex rules of inheritance or reuse
of the same class id, etc).
Thanks -Vincent
>> I have touched the CSS several times too and I know I was very
>> uncomfortable every time since I didn't know where what I was
>> touching was used and what were the rules governing what I was
>> touching.
>>
>> Maybe there are ways to write unit tests for CSS and if that
>> existed it would be even better than commenting I agree (even
>> though comments would still be needed to explain why something is
>> done in given manner).
>>
>> Thanks -Vincent
>>
>> On Jan 21, 2010, at 8:36 PM, Ecaterina Valica wrote:
>>
>>> If we want to have more comments in the CSS we should use a
>>> minimizer for the CSS, so that they are removed (along with
>>> the spaces) from the version we serve to the client. Some CSS
>>> files we have already do this.
>>>
>>> Another problem with CSS is that we add / modify some rules to
>>> assure compatibility with some specific browser. A solution
>>> would be to have specific files (just like the one we have for
>>> IE6) for the browsers that need those properties, so that we
>>> don't have to comment on every lines what compatibility is used
>>> for. Having multiple files that do the same thing can be hard
>>> to manage and sometimes browsers share behavior (depending on
>>> the engine). Ex: In the patch above the scrolling was only for
>>> Chrome and Safari.
>>>
>>> And the big problem in my opinion is that rules come in
>>> packages. To fix one bug you would need to comment a line in a
>>> class that is dependent to a line from another class, etc, etc.
>>> Ex: In the patch above the widths matter because of the
>>> float:left. Plus the cascading inheritance behavior of CSS.
>>>
>>> Right now we use comments in CSS to group properties, to
>>> specify overriding, to specify the browser that needs the
>>> property or to mark the hack.
--
Sergiu Dumitriu
http://purl.org/net/sergiu/