AviSynth 2.5.8 [Dec 30th] - Release! (was RC-5 [Dec 27th])

[Leak]

Originally Posted by Zep

It is an OS problem.

It's not an OS problem, as memory fragmentation only happens inside a program's address space - an x86 OS only hands out memory in pages (usually 4 kB in size) which don't have any fragmentation issues.

It's up to each application and/or library to support the memory management it wants/needs - but usually it's just malloc()/free() and that's that.

Java or C#/.NET CLR on any platform will garbage collect/compact memory without a hitch, but for C/C++ your program and all libraries it uses need support for doing that.

Usually it's just garbage collection and no memory compaction since that needs knowledge where the pointers in your objects are *AND* a mechanism to make sure nothing accesses the objects you're compacting while the compaction is being done...

But unless your OS is 100% managed (like MS's Singularity) it's up to the applications/runtime libraries to support all this, not the OS itself.

[josey_wells]

I recently pulled down V2.5.8 RC3 and tried to compile. I successfully got it to compile but there were >600 compiler warnings.

One could argue whether the secure versions of the library functions are really needed or whether some of the typecast are needed - IMO some of them are. But there were also a significant number of variable scoping problems.

Below is source code and diffs to sourceforge version.

http://rapidshare.com/files/14023423...2.5.8.rar.html

http://rapidshare.com/files/14023423...mpare.rar.html

Also, there does not seem to be a debug version of the devil.lib included.

Heopfully some of these changes will improve this already great product.

[Gavino]

Originally Posted by Leak

Java or C#/.NET CLR on any platform will garbage collect/compact memory without a hitch...

Garbage collection does have its downsides - mysterious bugs.

Long ago, I worked on an Algol68 compiler whose garbage collector started collecting itself. Talk about disappearing up your own backside...

[@IanB: Sorry for going off-topic]

[IanB]

The other part about running out of address space at 1.3GB usage is that Windows is a Virtual Memory environment. You really should be checking the amount of VM allocated to the process as well. And do not forget, in that 2Gb of available address space also lives the code, data and stack segments. 1.3GB is an incredible amount of memory. To get some context it is room for 170 high definition rgb32 frames or 2700 standard definition YV12 frames. The most temporally demanding filters may need upto 15 frames from the cache. Unless a given frame is requested more than once in a script the cache can offer no performance increase.

And talking about garbage collection and compaction environments here is off topic, this thread is about problems with version 2.5.8 Release Candidates, currently number 3. Please open new threads to discuss all these other interesting but off topic subjects.

Also release candidate means we are in Mortein/Baygon* the last few bugs just before we push the product out the door mode. We do not make radical changes like switch compilers or start using the "safe" libraries.


* - Bug spray products.

[josey_wells]

Originally Posted by IanB

Also release candidate means we are in Mortein/Baygon* the last few bugs just before we push the product out the door mode. We do not make radical changes like switch compilers or start using the "safe" libraries.


* - Bug spray products.

No problem, these can be queued for V2.5.9.

[IanB]

Changelist with respect to 2.5.8 RC3: See 1st post for full change list.

Additions:

* Updated Japanese documentation [20080813] (Niiyan).

Bugfixes:

* Fixed DirectShowSource() grey frames after end of stream + 30 frames.
* Fixed Histogram() Classic mode restore graph brightness. Regression.
* Fixed Compare() graph pixel values exceeding YUV limits.
* Fixed AddBorders() args negative value clamping.

Changes:

* Initial Memory Max value clamped to 512MB.
* Default Memory Max value restored to quarter of Free memory. Minimum 16Mb. As per 2.5.7.

As usual download from Sourceforge.
Enjoy!


If we do not find any show stopper problems this could become the official 2.5.8 release.

[Fizick]

IanB, I try translate changes.
What is difference: "Initial Memory Max value" (< 512 MB) versus "Default Memory Max value" (quarter of free, >16 MB)?
It is same parameter?

EDIT: So, what is (memory) changes "with respect to v2.57" ?

[IanB]

Originally Posted by Fizick

I try translate changes.
What is difference: "Initial Memory Max value" (< 512 MB) versus "Default Memory Max value" (quarter of free, >16 MB)?
It is same parameter?

Yes, they are the same thing.

EDIT: So, what is (memory) changes "with respect to v2.57"?

The "quarter of free, Minimum 16MB" is the same as 2.5.7 again.

The Maximum of 512MB is new, i.e. on a Vista64 with 4GB free 2.5.7 would start with a 1GB MemoryMax and could possibly crash out the 2GB address space limit, now 2.5.8rc4 limits to 512MB.

Earlier 2.5.8's used 50% of the free and crashed out the 2GB address space limit even easier.

All this just effects the Default or Initial MemoryMax value.

Users can still use SetMemoryMax() to set higher or lower limits to get themselves into trouble.

[Fizick]

IanB,
1. Thanks.
2. May be you (we) can somehow summarize some of the 2.5.7 to 2.5.8 changes for releasenotes.htm?
To provide the most important changes here.
Now it simply duplicates the changelist.htm
Similar to "DirectShowsource improvements,..."

Thoughts ?

3. Your sourceforge link is https instead of http

[LoRd_MuldeR]

Originally Posted by Fizick

2. May be you (we) can somehow summarize some of the 2.5.7 to 2.5.8 changes for releasenotes.htm?

http://forum.doom9.org/showpost.php?...22&postcount=1

[Wilbert]

That's not a summary

[IanB]

Originally Posted by Fizick

3. Your sourceforge link is https instead of http

Fixed! Damn those sourceforge https junkies.

Yes the release notes are a bit tatty but ...

[Fizick]

Ian, not related to summary, but please do not skip: I make some small changes (mostly typos of russian doc) at CVS

[IanB]

@Fizick,

So I take this as a request to make a 2.5.8 RC4a. I'll find some time latter this week.


Anybody else got some last minute doco/language type updates? Squeak now!

[Fizick]

I do not asked for it
But I ask somebody look to uptated description of ConditionalFilter (I do not use it much) at CVS http://avisynth2.cvs.sourceforge.net...?r1=1.6&r2=1.7 and wiki
http://avisynth.org/mediawiki/ConditionalFilter

[Wilbert]

What's the difference between a filter and a function in this case?

[Fizick]

IMO, AverageLuma(clip) and similar expressions with numeric result are not filters, but functions (in ConditionalFilter).
Filter must return filtered clip (in ScriptClip).

Previous text used "filter" and "function" in not consistant manner.
There were also some other non-clear fragments like this for ConditionalFilter:

The string filter can be any internal filter, but also some filters which are predefined (the Runtime Functions).

What internal filter may be alone used in ConditionalFilter?
Why internal and predefined only? What about plugin or user script functions?

The right changes are not obvious for me, it is hard topic for me, and proper English too, that is why I ask for discusssion.

[Gavino]

Originally Posted by Fizick

I ask somebody look to uptated description of ConditionalFilter (I do not use it much) at CVS http://avisynth2.cvs.sourceforge.net...?r1=1.6&r2=1.7 and wiki
http://avisynth.org/mediawiki/ConditionalFilter

I have taken the liberty to update the wiki. I hope this is OK.

This was before I saw Fizick's conversation with Wilbert, but I think that just confirms what the issue was, and I agree with Fizick.

[Fizick]

Gavino,
O.K.
I also added current_frame to list (see wiki).
Waiting for comments from Wilbert (and others), then shall update CVS.

[Gavino]

I have further updated the same wiki page to state explicitly that line breaks may be used in the ScriptClip string.