Andre Poenitz wrote:
On Wed, May 13, 2009 at 04:29:40PM +0200, Vincent van Ravesteijn - TNW
wrote:
In what ways were you using too much of it? I'm considering Qt as a
possible development environment to automate my business, so it
would
be good to know before I make the same mistake.
This was partly a joke, but had a real source. We were QFile to do various
filesystem-related operations in a cross-platform way, and it turned out that
this was, somewhat surprisingly, leading to gazillions of filesystem accesses
that were leading to slowness in some cases, e.g., when people were using
LyX to access files over slow networks.
Seemed like a good idea at the time, and in a way ought to have been.
Maybe, it's not fair to blame Qt. Maybe, the problem was how we use(d)
it.
There is a grain of truth in alomost everything. QFile[Info] did not get
the same amount of love in the Qt 3 -> 4 leap as other areas, leading to
a pile of fairly robust but not exactly fast code. There are prototype
implementations that speed things up considerably, but it's pretty
tricky to do that in a binarily compatible fashion...
And I think our problem had to do with using it for something it may not
really have been intended for, or at least not realizing that checking
to make sure two filenames really do point to the same file is going to
have to require filesystem access, if you're to be sure. And there was
some other oddity about how Qt handles non-existent files that makes
sense when you look into it, but isn't very intuitive.
Richard
