Help beta testing QuickLZ 1.40 with the new test framework

[Lasse Reinhold]

Hi all,

I've made a test framework which runs in IDLE_PRIORITY_CLASS, compressing, decompressing and verifying data of a given directory/drive.

If anyone has some idle CPU cycles and large amounts of data, you can contribute: http://www.quicklz.com/beta140.html

Source included.

[encode]

Thanks, will look at it!

[Simon Berger]

I didn?t get any error on much tested data.
But any logfile-output at the end of every candidate wouldn?t bad (mb of tested data, maybe any avarage percent of the ratio...)

[Nania Francesco]

Thanks Lasse!

[Lasse Reinhold]

Thanks for testing, Simon - will think about adding a log.

I'm thinking about adding a prize for finding a bug...

[Simon Berger]

do you really think there could be one?
The algorithm seems to be more or less simple (handles all data the same). I had many strange files in the tested folders.

[Black_Fox]

After 4 hours and 250 GB of data I found out only first executable was tested This will take a bit longer I guess...

[Lasse Reinhold]

Black_Fox, great - keep testing, I really appreciate it

Simon Berger: The algorithm is relatively simple, but look at LZF 3.0 (from 2007) which is even simpler and still had a bug. There are so many things that could go wrong:

a) regular data errors
b) spurious memory access, for example a read-access 1 byte beyond input buffer, which are dangerous because they normally go undetected until some random time.
c) specification errors according to manual, such as exceeding (original size) + 400 in the case for quicklz 1.40
d) strange behaviour which are not "bugs" such as being really really slow in certain siturations (in a development version I had a bug that made compression take minutes for just 1 kbyte of certain kinds of data), or suddenly losing compression ratio (for example, hash table getting corrupted). Many often forget to test "strange behaviour".

That was just a few things out of many. I have to test all this for all combinations of compression level, streaming buffer, etc, and on multiple operating systems and architectures.

[Stephan Busch]

Dear Lasse,

I have added you to the hall of fame because you created QuickLZ - please let me know if you allow me to do so. If you want to have a photo / birthday added, please let me know as well.


Yours,

Stephan

[Lasse Reinhold]

Hi Stephan,

Wow, cool, thanks alot

I have uploaded a picture at www.quicklz.com/orig.tif.

[Bulat Ziganshin]

btw, i've used for the same purpose freearc (and found 3 very rare errors in tornado, although it's much more complex algorithm. OTOH, quicklz is commercial library so i understand your need to highest level of testing). look at http://encode.ru/forums/index.php?ac...page=0#msg9695

i don't added there that fa allows to split data in independent solid blocks of given size, so you can test it not only file-wise