Wine FIXME Report January 2010

[Michael Stefaniuc]

On 02/09/2010 12:01 AM, Francois Gouget wrote:
> I'm pretty sure if I read the code I would understand this report better
> but I was not curious enough and unfortunately that means I'm pretty
> confused now.
You need to read only the SQL used to generate the tables ;)

> It seems the data is sliced three ways:
I needed some catchy titles; I know those aren't that great. If somebody 
has better ideas I'm all ears.

>   * The Most Popular Messages
>     - Based on the name it sounds like someone voted on them, but then
>       the definition says they 'are the most prevalent' which sounds like
>       they are just the most frequent ones.
>     - So I look at the number of lines where they show up and the number
>       one shows up in just 119 lines, while some way down it seems there
>       are some that show up in hundreds of thousands of lines. So I'm
>       lost.
With 'prevalent' I mean the number of files that contain that message
not how often those show up in that message. Basically the equivalent of
grep -l | wc -l and sorted by that number.

Reason:
   Wine is too noisy and people tend to ignore the "stray" fixme and err
   messages here and there. They notice only if a message floods the
   output. Then it gets silenced to print only once and the message
   disappears from the radar. This tables tries to show the fixme/err
   messages that a lot of people see and overlook. As a lot of people see
   them they are "popular". Like the fixme("stub") messages in
   DllCanUnloadNow() that got fixed aka removed because of this report;
   everybody was seeing and ignoring them.

>     - It also does not help that the order of the Files and Lines column
>       is reversed compared to all the other sections.
I did it to make it visible that the table is sorted by the first column
aka "Files". Probably dropping the "Lines" column altogether would have 
made it clearer.


>   * Noisy Popular Messages / Functions
>     - So here the definition says they 'show up at least in 1% of the
>       collected reports', so it sounds like it's the messages/functions
>       that impact the most reports. So it's most frequent per report
>       count.
Not quite. The "The Most Popular Messages" impact the most reports. I
came up with this reports after the "The Top Ten Single Charts" turned
out to be pretty useless. There are messages that repeat hundreds of
thousands of times in *one* single file. And there are no other files
out of ~2500 files with that message. While that is annoying for the guy
that got that flood, it is statistically irrelevant. Might be his setup;
might be a broken commit round, who knows. But those are definitely not
worth silencing aka using the if(once++) fixme() trick.

Thus I tried to find fixme/err messages that are noisy (lots of repeated
messages in a file) and relevant (1% of the files). I figure Alexandre
might be willing to accept the if(once++) fixme() trick for those.

>     - They are still sorted by Line count though. Shouldn't they be
>       sorted some other way?
I have thought about sorting by the average number of messages per file;
I'll look at it tomorrow and see if it makes more sense.

>     - Is there a difference between a file and a report?
No, not in this context. Though I don't like either of those names:
- A file is also an email that has no fixme/err/warn messages.
- A report is conflicting with the "Wine FIXME Report".
But I couldn't come up with a better name back then. A "log" should be a
better name for the files that contain fixme/err/warn messages. But I'm
open for other names too.


>   * The Top Ten Single Charts
>     - These really look like they are the most frequent by line count
>       (based on the line count). But I have some doubts.
This are the maximum hits per file and not a sum of all hits. Pretty
useless and I've thought about dropping this tables as they aren't
statistically relevant; more than 50% of those messages show up in only
one file.


bye
	michael