Memory leaks in C API

[Frederik Gladhorn]

Torsdag 8. mai 2014 10.31.32 skrev Jarek Czekalski:
> Hi Frederik
> 
> The patch nr 2 is definitely something good, though I didn't analyze it
> fully yet. However regarding patches 1 and 3, I have doubts.
> 
> xmalloc and xfree could potentially be used to:
> 1. insert breakpoints
> 2. output debug info
That you allocated memory? Great. I'd suggest using proper tools for that.

> 3. using different memory management functions
Sure, re-introducing them will be trivial if someone feels the need. How much 
development has this C library seen lately? The largest change recently was 
white-space cleanup in 2011.

> 4. introducing additional code monitoring or catching memory leaks

1 and 4: You can do that today. You can also LD_PRELOAD and use address-
sanitizer and valgrind.


> Patches 1,3 would remove this possibility. In general, if something is not
> harmful I would respect original author's ideas.
I'd prefer simplicity and maintainability. Reintroduce it when needed.

> 
> Note that I am just a beholder and have no powers in speechd. However I'm
> happy you investigate speech dispatcher api, because recently Google guys
> reported general nature problems with speech dispatcher [1]. To the extent
> that they disabled speech dispatcher support by default. My long term plans
> include getting deeper into this.

I'm not surprised. I didn't look at the actual code yet, only the C API which 
is one tiny file.

Cheers,
Frederik


> 
> Jarek
> 
> [1]
> https://groups.google.com/d/msg/axs-chrome-discuss/PnmzZdvM0Bs/WrAxsN5slEkJ
> 
> W dniu 2014-05-08 09:57, Frederik Gladhorn pisze:
> Hi,
>  
> I was just playing with the C API of speech
>         dispatcher.
>  
> The first patch removes xfree which doesn't
>         make much sense, it did a nullptr check before calling free.
> The second patch fixes a few memory leaks.
>         Note that it adds a function to actually free a connection, that
>         was pretty hard work so far.
>  
> The third patch I leave up to your judgement,
>         it simplifies the code by also getting rid of xmalloc which is
>         malloc with a warning and exit if we're out of memory. Imho
>         there is no point of handling that and it clutters the code.
>         It's quite unlikely that the tiny C client API can do anything
>         in case of OOM.
>  
> Greetings,
> Frederik
> 
> 
> _______________________________________________
> Speechd mailing list
> Speechd at lists.freebsoft.org
> http://lists.freebsoft.org/mailman/listinfo/speechd