String concatenation should be replaced with join()

[stoarca]

Here are the results of a modified spitfire test (https://code.google.com/p/spitfire/source/browse/trunk/tests/perf/bigtable.py) on my machine using

def run_test(t):
  xhpy.pylib.core.ENABLE_VALIDATION = False
  table = <table/>
  for row in t:
    tr = <tr/>
    for col in row.itervalues():
      tr.appendChild(<td>{col}</td>)
    table.appendChild(tr)
  return str(table)

1000x10 table (the default):

Mako Template                                  23.93 ms
StringIO                                       31.38 ms
cStringIO                                       5.48 ms
list concat                                    11.71 ms
xhpy                                          120.58 ms

10x10000 table:

Mako Template                                 247.07 ms
StringIO                                      310.89 ms
cStringIO                                      57.64 ms
list concat                                    44.02 ms
xhpy                                         5738.22 ms

10x100000 table:

Mako Template                                2393.53 ms
StringIO                                     3110.47 ms
cStringIO                                     558.29 ms
list concat                                   409.80 ms
xhpy                                      1733126.09 ms

While the last one might not be a totally realistic example, the first two are, and 5 seconds is not a reasonable render time. I can submit a pull request if you let me know how to use the pdb within xhpy elements. It skips most things for some reason.

Coincidentally, I think you may have been my frosh leader in 2008. I started working on the same thing for a couple of days before I found yours. Great work!

[candu]

Apologies for the delayed reply - things have been insanely busy lately on the work/life front.

This is a good point. I haven't put much effort into optimizing XHPy, so there's bound to be some low-hanging fruit around.

Using join() shaves ~50% of rendering time in the 10x10000 case. From a quick profiling run, the rest of the overhead is more spread out: init() for all those document nodes, instanceof as it collapses :x:element instances into strings...

On Wed, Sep 12, 2012 at 4:14 PM, stoarca [email protected] wrote:

Here are the results of a modified spitfire test ( https://code.google.com/p/spitfire/source/browse/trunk/tests/perf/bigtable.py) on my machine using

def run_test(t): xhpy.pylib.core.ENABLE_VALIDATION = False table =

for row in t: tr = for col in row.itervalues(): tr.appendChild({col}) table.appendChild(tr) return str(table)

1000x10 table (the default):

Mako Template 23.93 ms StringIO 31.38 ms cStringIO 5.48 ms list concat 11.71 ms xhpy 120.58 ms

10x10000 table:

Mako Template 247.07 ms StringIO 310.89 ms cStringIO 57.64 ms list concat 44.02 ms xhpy 5738.22 ms

10x100000 table:

Mako Template 2393.53 ms StringIO 3110.47 ms cStringIO 558.29 ms list concat 409.80 ms xhpy 1733126.09 ms

While the last one might not be a totally realistic example, the first two are, and 5 seconds is not a reasonable render time. I can submit a pull request if you let me know how to use the pdb within xhpy elements. It skips most things for some reason.

Coincidentally, I think you may have been my frosh leader in 2008. I started working on the same thing for a couple of days before I found yours. Great work!

— Reply to this email directly or view it on GitHubhttps://github.com/candu/xhpy/issues/4.