Re: Issue 1211 in lilypond: [PATCH]Optimizations for pure-height approximations. (issue1817045)

[Graham Percival]

On Sat, Nov 20, 2010 at 9:44 PM, Neil Puttock <address@hidden> wrote:
> On 20 November 2010 00:22, Graham Percival <address@hidden> wrote:
>
>> That *is* the "big picture".  If the "time" field is fixed, and
>
> I've just looked at the "time" field issue, and it seems to me it's
> deliberately disabled.  The timing info is switched on via the
> `dump-cpu-profile' option, which says it's system dependent: it likely
> produces misleading results.

1) is it likely to produce misleading results using the same computer,
same operating system, same build environment?
if yes... and if the range of "misleading values" is too large to be
statistically significant... then fine, let's drop the idea.  If the
"misleading values" produce a spread of, say, 10%, then it would still
be useful to catch huge drops in speed.
if no, then this would be useful for the "make check" stuff.  I think
this is the more likely option.

2) is it likely to produce misleading results using the same computer,
same operating system, but possibly changed build environment?
if yes, then we'll document that the official regtest times may be misleading.
if no, then we'll document that the official regtest times are
generally ok, but occasionally they'll get all wacked out.  I can tell
people exactly when I've changed the build environment (i.e. new OS,
or doing a GUB git pull), so we'd know when the results should be
disregarded.

Cheers,
- Graham