Re: [bug-gperf] gperf as a library

[Bruno Haible]

Hi Mark,

> I had one suggestion about usability and four suggestions about performance.

Only one issue per mail thread, please! Otherwise discussions become
unmanageable and/or are not goal-directed.

> USABILITY SUGGESTION ( VERY INVOLVED and/or DIFFICULT )
> 
>   I would like to have a limited version of gperf which could be invoked
> while running C or C++. I don't know how to do this,
> but I understand what possible syntax might be.
> 
>    struct gperf
>    {
>       unsigned long MAX_WORD_LENGTH;
>       unsigned long MIN_WORD_LENGTH;
>       unsigned long MAX_HASH_VALUE:
>       unsigned long MIN_HASH_VALUE;
>       unsigned char * alloc_array;
>       unsigned long * shifts; // 0 and 1 in the the perm2.gperf in
> examples directory
>       char ** words; // to borrow, but not own, an array.
>   };
> 
>  char *the[] = { "hi", "there",""}; // I don't know if this is
> syntactically correct.
> // The "" acts as a "end of list" identifier
>  gperf x;
>   initialize_gperf(&x,the);
>   unsigned int n1 = hash( "there", 5, x);
>   unsigned int n2 = hash("dude",4, x);
>   assert( n1 ); // found
>   assert( ! n2 ); // not found
>   destroy_gperf(x); // freeing alloc_array  and shifts, but not words

What does the 'gperf' type contain?

If it contains just the arrays, you lose at least a factor of 2
through the "interpreter" that does the computations according to
the values in the arrays.

If it contains machine code, then you need to link with GNU lightning,
libgccjit, or LLVM - and then you have linked in a lot of libraries
and accepted platform constraints. Not to forget the system calls needed
for putting the generated code into executable memory. Is that worth it?

Bruno