Re: configuration patch - final version

[Martin Pala]

Jan-Henrik Haukeland wrote:

> Elaborations (I'm on windows right now, and cannot try, this is more thinking
> aloud)
>
>
> Write:
>
>  1) first write the timestamp:
>
>  time_t now;
>
>  fwrite(time(&now), 1, sizeof (time_t), binaryfile);
>
>  2) Then the length of the process list
>
>     int length= get_process_list_length();
>
>  fwrite(length, 1, sizeof (int), binaryfile);
>
>  2) Then the list:
>
>  for(p= processlist; p; p= p->next)
>         fwrite(p, 1, sizeof *(p), binaryfile);
>
>
> Read:
>
> 1) Get state file timestamp
>
>   time_t stamp;
>
>   fread(now, 1, sizeof (time_t), binaryfile);
>
> 2) Get length
>
>   int length;
>   fread(length, 1, sizeof (int), binaryfile);
>
> Process_T new;
> Process_T temp= NEW(temp);
>
> for(i=0; i<length; i++) {
>
> if(fread(temp, 1, sizeof *(temp), binaryfile) != sizeof *(temp))
>   error_exit;
> }
> if((new= get_process(temp->name))
>   copy temp.do_validate to new.do_validate;
> }
>
>
> Simple and effective (if it works that is :)
>
>
> Jan-Henrik
>
>
>  
Sounds good. The BER rules uses the same princip - it defines tag for 
each data type + length + data part length. There's also support for 
structures, etc. (primitive and constructed types, support for 
user-defined data type classes),  is possible to validate the stored 
data in the parser as well. Problem is, that it is little bit more 
complicated. We can use some parser that is ready for use (for example 
OpenLDAP is distributed with libber, but i don't like that parser - when 
i went through this code i found few "not-very-nice" chunks of code).