On 20 December 2013 14:09, Timo Teras timo.teras@iki.fi wrote:
On Fri, 20 Dec 2013 13:56:08 +0000 Peter Dunkley peter.dunkley@crocodile-rcs.com wrote:
One of the reasons I used libunistring for this detection is that for pretty much all of the code fragments I found online for doing this in a "simple" way I saw people pointing out flaws in those algorithms. Can you confirm that this code doesn't have any of those flaws and is guaranteed to work in all cases (has this implementation been stubbed out and tested fully by someone here)?
Did you read the document on the URL it refers to? It is quite thorough explanation of what it does, it's correctness and speed. It also explains that the motivation for implementing it was because all those snippets in the internet are seriously flawed.
I did see the document. It looks good. Does you have first hand
experience of whether that code is valid and correct or not?
I personally tend to trust a release library from GNU somewhat more than code on a web-site - however good that web-site and documentation looks.
Does this really improve performance? Only a tiny, tiny, subset of
libunistring is used. As a result it doesn't really matter if libunistring in general is slow, just whether or not the one function used from libunistring is slow.
I'm referring specifically to the function you use. Please check the URL for performance comparison. While libunistring is not benchmarked separately, one can see with 0.1 second look at libunistring's implementation that it will be slower in performance and is likely something close to iconv()'s implementation.
I don't have any objection to this change as long as you are 100% sure that the algorithm from that web-page works correctly. The algorithm looks good, the web-page looks good, the documentation looks good, but I'd really prefer it if the implementation was explicitly and fully tested before the libunistring call is replaced.
Something as simple as a loop through all possible values calling your function, the libunistring function, and comparing the results would be perfect.
Regards,
Peter