Zugg Software :: View topic - [CMUD] Speed optimization hints and rules.

Novice Joined: 17 Dec 2007 Posts: 49

General feeling since using cmud after years in zmud is... a bit faster!
But.. ctrl+q test says muuuch sloower. (5,0 zmud to 15,0 cmud.)
Always tried to optimize my scripts but it was always based on unprooven users hints or general programming rules.

Post here your hints about script speed optimization.
Prooved hints please and clear discussion.
Maybe this should go into FAQ section some day.
Any1 with better forum experience plz keep discussion ;)

Novice Joined: 17 Dec 2007 Posts: 49

To better understand speed issues can any1 explain how trigger matching is made? Or point to relative topic ?

Novice Joined: 17 Dec 2007 Posts: 49

Triggers (no variable write):

Posted: Tue Apr 29, 2008 12:45 am

Trigger matching just uses the regex engine - you can see the regex that's being used for your pattern on the Test Pattern tab. There're many guides, tutorials, articles and tools out there that explain how regex matching works and give tips for optimising them.

If you're unsure about which option is going to be faster, the easiest way is to use the untitled session and write a script that'll create a couple of hundred triggers of each type. Then disable all the options but one and see how long it takes. Repeat for each option.

A couple of things you need to know about speed testing:
1) It'll be slower if the buffer isn't full, because CMUD needs to allocate memory to hold the lines in the buffer. Press Ctrl+Q a few times to fill up the buffer before you start speed testing.
2) It'll be slower the first time you run each trigger in a session, because CMUD needs to compile them first. Again, press Ctrl+Q before you start speed testing to give you an accurate result.

Most importantly, though, is that speed depends on many things. You should look into optimising your scripts as well as your patterns using local variables, events and other features that've been added to CMUD.

Posted: Tue Apr 29, 2008 12:53 am

Also, tell us what version of Windows you are using. Ctrl-Q speed is significantly slower on Vista for example.

Trigger matching is done using the industry standard PCRE library (Perl-Compatible Regular Expressions). However, with CMUD's compiler and caching, there are many different optimizations, all at too low of a level to really describe here.

Honestly, the best way to answer your specific examples is to try them in a blank CMUD session and use the Ctrl-Q test to compare speeds. But very often you will find that it will depend upon the exact text you are trying to match.

In *general*, using a single trigger is faster than multiple triggers (your first example). Btw, it would help if you numbered your examples to make them easier to refer to.

CMUD used regular expressions (PCRE) internally. PCRE doesn't have any knowledge of stuff like CMUD variables. So CMUD performs a conversion step before calling PCRE. For example, when using a @variable in a trigger pattern, CMUD expands the variable before sending it to the PCRE library. CMUD caches this result and continues to use the cached value until the @variable changes value. So it doesn't actually matter whether you use a variable of whether you put all of the text into the trigger verbatim. The same end-result gets sent to the PCRE library.

Using &varname in a pattern is really the same as using "var=%1" in the value. Again, PCRE doesn't know about CMUD variables. So when CMUD converts this pattern to a regular expression, the &varname is replaced with (.*) and the variable name is stored for later. If the pattern matches, then CMUD internally does the "var=%1" to assign the command.

None of the things that you mentioned should really have much difference between CMUD and zMUD...both use the PCRE library. A good tool to look for speed issues in CMUD is to open the Script Debugger in the Window menu. This will slow down your scripts, but you'll get to see exactly what triggers are firing and what scripts are running, along with timestamp values so you can see what scripts are slow.

Another good tip is to put your various triggers into class folders, and then disable the class folders and do the Ctrl-Q test to determine where your slowest triggers are. Once you identify the triggers that are the slowest, then you can ask about them specifically for optimization.

Finally, try to avoid putting a * at the beginning or end of a pattern. CMUD already does this by default (unless you anchor it with a ^ or $). So adding your own * just makes PCRE take longer (since now it has two wildcards together).

Novice Joined: 17 Dec 2007 Posts: 49

I understand, that ctrl+q test measures JUST matching speed, not result of matching. It doesnt fire any triggers ?
So point of my question is only matching purposes of any optimizations.

Not everything goes with regex i think. For example how about nesting triggers deeper inside class folder tree... have read somewhere its much slower.

Novice Joined: 17 Dec 2007 Posts: 49

So there is no real speed difference in:

Posted: Tue Apr 29, 2008 2:51 am

Novice Joined: 17 Dec 2007 Posts: 49

Wizard Joined: 25 Mar 2003 Posts: 1113 Location: USA

Magician Joined: 17 Apr 2005 Posts: 496

Atomic grouping is what Larkin's talking about, it's pretty interesting stuff.