Re: dynbuf: PR #5300

On 4/30/20 3:10 PM, Daniel Stenberg via curl-library wrote:
> Hi team,
>
> In PR 5300 I'm introducing a library-wide change that I wanted to make
> you all aware of!
>
>
> # Feedback?

I think there is also the sequence at
https://github.com/curl/curl/blob/master/packages/OS400/ccsidcurl.c#L855
that could be transformed, although array elements are not characters.
Other realloc calls in OS400 code only shrink buffers to spare memory.

Suggestions:

- Implement auto shrink when shortening data under an arbitrary
watermark, for example 20% of the allocated memory.

- A function to grow allocation without initializing it: something like
(conceptually) accepting Curl_dyn_addn(&dbuf, NULL, size), maybe
returning a pointer to the extended memory block. The motivation is: in
some circonstances, OS400 does not like moving pointers as characters;
if the dynamic array element contains a pointer, it may lose its
addressing capability after byte to byte copy, thus a direct access to
the allocated block is required for proper initialisation. This would
also be useful to allocate a buffer when the final data size is unknown
(before a read, for example).

- Curl_dyn_head(): opposite of Curl_dyn_tail(), keeping the leading
bytes. Although very simple to perform without such a function, it'll
add implementation details abstraction.

- In debug mode, store DYNINIT also after the data and check it: this
may detect buffer overflows. This could also be done before the data,
but beware of alignement.

-------------------------------------------------------------------
Unsubscribe: https://cool.haxx.se/list/listinfo/curl-library
Etiquette: https://curl.haxx.se/mail/etiquette.html
Received on 2020-05-01