Re: [boost] [proposal] Sequence container with O(log n) randomaccess _and_ O(log n) insertion/removal (wherever)

From: Jason Hise (0xchaos_at_[hidden])
Date: 2006-10-30 19:18:16

I have started looking through the code.  It looks like some pretty
good work thus far.  My first comments:

A minor nit: you probably shouldn't be using _STL_ as part of your
include guards... it makes it look as though the library is already a
part of the stl (and it isn't... yet ;) )

Instead of defining your own greater than and less than functors, it
may be preferable to use greater and less, defined in the <functional>
header.  Just like yours, the stl greater requires Ts to be less than
comparable rather than greater than comparable.

Your static allocator is probably a bad idea... if someone creates a
static instance of this container type, the allocator could be
destroyed before the container destructor is invoked.  The container
destructor would then reference the non-existant object.  There are
ways around this, but IMO it would just be better practice to have
each container own its allocator.

You use c-style casts from literals to type W, which should either be
changed to C++ style casts or explicit constructor calls.

It would be good to find a way to rewrite your try-catch blocks in
terms of helper commit or rollback objects, which release resources in
their destructors unless cancelled (after a successful operation,
calling cancel on such an object would be the equivalent of commit).
This way, the code will still compile if the end user chooses to
disable exception handling, or if the compiler does not support it.

The header is pretty large... breaking it into smaller files, even if
these aren't publicly exposed, might help readability and
maintainability.

The commenting is good, but nothing is a substitute for formal library
documentation.  At the very least, an html or pdf file should be
provided.  The most important things to cover are which concepts are
implemented by your container and iterators and what the complexity
guarantees of all operations are.  A discussion of the implementation
or links to online resources about it would be nice as well, but not
as vital as the first two items.

-Jason