requesting feedback: memory management in flex

[John W. Millaway]

One of the items on the TODO list for flex is to allow the programmer to
specify exactly HOW and WHEN flex allocates memory.

Currently, flex allocates a 16k+40b  buffer in the first call to yylex(), then
realloc's memory as needed, by doubling the size of the buffer each time.  On
the first call to yy_push_state, flex allocates 25*sizeof(int) bytes for the
stack when using scanner states.  The stack is realloc'd as needed in constant
chunks of 25*sizeof(int).

There is no documented API  to free the 16k+ buffer in plain flex  (but you can
free it in the C++ scanners and reentrant C scanners.)

I'm requesting suggestions for things you would like to do with flex, but that
are not currently easy to do because of the way flex handles memory.

The obvious ones are:

* Ability to override the amount of memory alloc'd and realloc'd.
* Ability to enforce an upper bound on the buffer sizes, with error handling.
* Ability to override the default alloc/realloc/free functions.
* Ability to free the 16k+ input buffer in traditional flex (non-reentrant C
scanner).

Any other suggestions?



__________________________________________________
Do You Yahoo!?
Yahoo! Movies - coverage of the 74th Academy Awards®
http://movies.yahoo.com/