Hi,
I have finished work on the improved stringlist. There are few changes
in the public section: A new public property OnCompareItems:
TStringListSortCompare allows to define the compare function, while
keeping sslAuto as sortstyle. If OnCompareItems uses data from the
object instead of or in addition to the string itself for sorting, the
new public boolean property ObjectSort must be set to true; in that
case, the function Indexof will work like for unsorted strings. Finally,
the find method can be called with an additional object as parameter, to
allow finding of string-object combinations through binary search. I
made a few minor improvements, too, so there is no speed penalty for the
increased flexibiity.
But before finalizing the patch and submitting it to the bugtracker, I
have some questions.
1. It is not allowed to change an element of a sorted stringlist or
insert a new string at some specific index. If that happens,
SSortedListError is signaled. Shouldn't that also be the case when
ExchangeItems is called on a sorted stringlist? I suggest adding the
error message there too.
2. Eventually, I can also address the randomization problem by making a
non-random choice of the Pivot Element according to
/On Fri, Nov 30, 2018 at 2:29 PM Andrea Mauri
//<andrea.mauri...@gmail.com>//wrote: /
/if TStringList sort will be changed, please take into account the
issue related to randomised pivot initialisation in TStringList sort
and the consequences to Random sequence (//http://forum.lazarus-ide.org/index.php?topic=43257.0//). /
/On Fri, Nov 30, 2018 at 2:29 PM Andrea Mauri
//<andrea.mauri...@gmail.com>//wrote: /
/if TStringList sort will be changed, please take into account the
issue related to randomised pivot initialisation in TStringList sort
and the consequences to Random sequence (//http://forum.lazarus-ide.org/index.php?topic=43257.0//). /
/https://en.wikipedia.org/wiki/Quicksort#Choice_of_pivot//suggest to either use a random value as pivot or the mean of the first,
middle and last element. Bart /
However, I am sceptical whether that would be worth the overhead -
whatever the overhead is, because it is absolutely unclear to me what is
meant with "the mean of the first, middle and last element". You can
caculate the mean for numbers, but not for strings. So I would rather
leave these things as they are now.
But maybe there is an easy way to save the state of the random number
generator routine before sorting and to restore it when the sort is done?
3. The documentation should be updated. Is there a documentation about
how to proceed to update the documentation for a class? There are a lot
of different places with descriptions and explanations, I have no idea
what to update and how to do that.
4. Are there test cases to validate the new stringlist, to make sure no
existing code is broken? I have tested everything I could think of, but
there may be constellations that I don't imagine. As stringlists have
certainly been thoroughly tested when they were first written, there may
already exist some "test suite" to test the behaviour. That would be
very helpful and I would be glad to run additional tests before
submitting the patch.
Franz
_______________________________________________
fpc-devel maillist - fpc-devel@lists.freepascal.org
http://lists.freepascal.org/cgi-bin/mailman/listinfo/fpc-devel
