[Ur] request for review of file save ffi extension

[Adam Chlipala]

This code looks basically OK to me.  I don't think it maintains the true 
transactional semantics, but you may be OK with that.  For example, if 
multiple transactions are writing multiple files, and some filenames are 
being written by multiple transactions, then disk states can be observed 
that aren't consistent with some serial order of transactions.

The commit handler doesn't seem to close the file handle in exceptional 
cases, which probably leads to a resource leak.

Minor point: I suggest strdup() as a faster way to make a string copy, 
vs. the malloc() and strcpy() you have now.

On 10/25/2015 10:34 AM, Orchidaceae Phalaenopsis wrote:
> Hello,
>
> I've modified the file upload demo slightly to save the file to the
> hard disk instead of echoing it back:
>
> * https://github.com/orchid-hybrid/urweb-file/tree/master
>
> In particular I have used ffi handlers to make saving the file a
> transactional operation. I've been very careful to follow the manual
> on correct use of the ffi but I would really appreciate if people
> would look over this and comment on whether it's correct or not:
>
> * https://github.com/orchid-hybrid/urweb-file/blob/master/file.c
>
> I'm worried about not freeing memory that needed to be, double frees,
> using memory after free, things like that.
>
> (N.B. This is step 1 of a larger project, I need to save files to a
> folder so that they can be processed by some unix tools before the
> results are published)
>
> Thank you very much!