Rx wrote:
> For example, if the user doesn't have an allocated buffer yet but has
> strict memory constraints, they may need to know how many groups there
> are before allocating the buffer. They could pass NULL as grouplist and
> maxcount as an upper bound for the buffer size. Then, when the function
> returns the count, it would provide a size that is guaranteed to fit
> within the user's memory constraints.
This is the style that many Windows APIs use. I hate this style, because
* As a programmer, I have to write 2 calls to a certain function.
* The kernel has to collect the information twice.
* The size is still *not* guaranteed to fit, because there is a race
condition: The sysadmin might add the user into another group,
just between the two calls.
I much prefer the approach with a stack-allocated buffer of small size.
See e.g.
gnulib/lib/xgetaname-impl.h
gnulib/lib/string-buffer.h
It uses
* multiple calls to the function only rarely,
* thus is efficient (on average),
* and does not have race conditions.
Bruno
