Re: [PATCH 00/19] Add hash table to gdbsupport

[Tom Tromey <tom@tromey.com>]

Simon> I spent a bit of time reading the interface of your hash table, and that
Simon> was a point I found unfortunate.  The type V must have a default
Simon> constructor and an "empty" state (I guess that's what you mean by
Simon> sentinel), even if it doesn't make sense for the rest of the program.

It's maybe worth noting that this is not a step backward from htab_t,
which only stores pointers.

Simon> How do other implementations (of open addessing hash tables) typically
Simon> deal with this?  From what I recally, other implementations I have used
Simon> in the past didn't have this requirement.

I don't know in general.  I looked at one and it uses a separate vector
of flag bytes to indicate which values are valid.

Simon> And that makes me think that a question I did not see answered is: I
Simon> don't want to undermine your work, but what is the rationale for
Simon> implementing it ourselves?

In my view, integrating an external C++ package like this is pain both
legally (at least, you have to find one with the appropriate license)
and technically.  Even just trying to integrate the GCC hash table was a
pain.  The typical problem, IMO, is that it's unusual to write a
standalone class like this, instead one introduces dependencies on a
variety of other things.

Tom