In my last post I revealed my fix for the hg annotate loading issues. My fix reduced the loading time to a relatively reasonable ~8sec considering the fact that currently the loading time is ~20sec. However, it was still bugging me that Mat’s implementation was producing faster loading times than my fix. His implementation is ~1sec faster.

Mat’s Implementation

The problem is that this implementation doesn’t use valid HTML. Mat uses two non-standard tags, x and l#parity#. Although, I must say that this is a very unorthodox and smart solution to this problem:

annotateline = '<l#parity#><x><a href="#url#diff/#node|short#/#file|urlescape#{sessionvars%urlparameter}">#author|obfuscate#@#rev#</a></x><a href="##lineid#" id="#lineid#">#linenumber#</a>#line|escape#</l#parity#>'

Altering Mat’s Code into Valid HTML

As an experiment I decided to test what would happen if I took Mat’s code and replaced the x and l#parity# tags with divs. Would the implemenation still remain as fast? Would it still be faster than mine? The following is the altered code:

annotateline = '<div class="l#parity#"><div class="codeauthor"><a href="#url#diff/#node|short#/#file|urlescape#{sessionvars%urlparameter}">#author|obfuscate#@#rev#</a></div><a class="codeline" href="##lineid#">#linenumber#</a>#line|escape#</div>'

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28

div.codeauthor { 
    display:inline-block; 
    width:16ch; 
    text-align: right; 
    color:#999999; 
    text-decoration:none;
    margin-right: 25em;
} 
a.codeline { 
    color:#999999; 
    text-decoration:none; 
    margin:0 10px; 
}                                   
div.l0 {
    background-color:#f6f6f0;
}                                                                    
div.l0, div.l1 { 
    display:block; 
}
pre.completecodeline { 
    font-size:12px; 
    line-height:1.4em; 
}
a.codeline:hover,
a.codeline:visited,
a.codeline:active {
    color: #880000;
}

Results

I tried testing the altered code and I was getting a loading time of ~7sec for this cpp file. However, I was never able to get into the ~5sec region which Mat’s original code was sometimes able to achieve. This altered version was only able to reach a minimum time of ~6sec.

The point is that this altered version is a bit faster than my fix but slower than Mat’s original implementation. I compared the file sizes of the 3 versions:

• My fix: 2.4MB
• Mat’s Fix:1.8MB
• Altered version of Mat’s Fix: 2.2MB

It is interesting to note that using div tags instead of an x and l#parity# tag increases the file size by 18%. I don’t know why that happens but somehow the x and l#parity# tags are more efficient than div tags. Nonetheless the reasons don’t matter, the altered version of Mat’s fix seems to be the best solution to this problem at this time.