On my article on Caching Guidelines for MODX Revolution, Marc Hinse asked a very good question that I don't want to leave unanswered.. however the answer is slightly longer than a simple comment, so this follow-up tries to go into what he's asking.

This was his comment:

A very good question, and now that the basic caching has been cleared up, I suppose we can move on and dive a bit deeper. The source-order recursive parsing makes it fairly complicated to explain (and I'm not sure if I know all the ins and outs myself either), but I'll give it a shot.

Parser Basics & Tag Order

From what I know, the parser collects all tags and starts executing them from start to finish, and inside out - so first the tag nested all the way down, and the outer most tags last. Everytime it has processed something, it would check that output for any tags and execute those - again "inside out". This is for all cached tags. It waits with processing uncached tags until all cached tags have been executed. So you could say this is the parsing order:

  1. Nested cached tag
  2. Cached tag
  3. Nested uncached tag
  4. Uncached tag
To make it all make more sense, I like to envision it that tags are put in some sort of queue, with the priority being set to any of the 1-4 above where 1 is top priority, and 4 is lowest priority. It's also sorted on source order, so a tag that appears in the header is executed before a tag in the footer - also based on the priority.

Furthermore it keeps this queue and parsing order in different levels. When a tag is found and executed its return value is immediately checked for tags, which are then added to a queue and executed. Any return value of those tags will be checked for tags which are added to a queue and executed. Any return valueof those tags.. well, you get the idea right? It's completely recursive and the parsing order is in effect per level (or scope, or context, whatever you want to call that).

Parsing Tags

Everytime a tag is parsed, the output will be checked to see if there are any tags in there, and if it finds any, they are added to the queue according to their priority and source order. Then it goes back to that queue and executes whatever is left based on its priority and source order.

When all cached tags have been processed, the result of all that (so with all uncached tags in place) is written to the cache for future requests. In other words - any uncached tags will appear as-is in the cache file, and will be picked up and executed on every request.

In Marc's example there's another layer of complexity added with the output modifiers. I think it first looks out for "inner tags", so in your example both [[$chunk1]] and [[$chunk2]] would be parsed before [[*tv]]. In that scenario it could be the following after one round of parsing:

Then in the next "round" of finding tags and adding it to the queue, it would parse the chunks to return your uncached snippet call and to replace the pagetitle tag with the pagetitle. So by now, we have the following:

Now we see that we have a cached TV tag, with uncached snippets in there. We know that cached take priority in the queue, so at this stage we would see the output modifier being executed (as well as other cached calls of course), resulting in an output of (assuming tv = 1):

At that point the cache would be written to file and parsed on every subsequent request. So even when nesting, as long as the outer elements are cached it should only save uncached elements for parsing on every request. This also is true when using an uncached placeholder in a snippet tpl chunk - they would be left in the rest of the output when writing to cache, for processing on every request.

Parsing Order & Conditional Logic

Basically, if you have conditional logic using output filters (or the If snippet for that matter), the "inner" tags will be executed before the outer (conditional) tags. What that means is that when you have the following example:

While this seems fine, it will parse inside-out, meaning that the FirstChildRedirect snippet will always be executed, and you end up being redirected to the first child, and the conditional logic is moot.

Now, knowing that we can nest tags anywhere, we can trick the parser somewhat, and put another pair of double brackets around the conditional logic, and remove them from within the conditional logic:

This will execute the conditional logic first, and will return either "!FirstChildRedirect" or "$video-gallery", which, thanks to the outer pair of two brackets, will form a valid tag which will then get executed after the conditional logic takes place.

This doesn't just stop at the name of the tag to load though, here's an example (from Jason "opengeek" Coward) illustrating just what you could do know that you learned this trick.. First we have a chunk (let's call it "sidebox"), which has the following contents:

And how it's being used (again, assuming the name "sidebox"):

Here's an article to read if you're new to chunk parameters and placeholders, but if you know that - and now know this trick - you may get the feeling there's some powerful applications for stufff like that..

Notes

This is the caching as I understand it at this point of time... There's probably some subtleties missing, but beyond that I'm quite confident it is correct and all. ;) I'm glad that Jason Corward was willing to give this article a read before hitting the final "Publish" button though, and pointing out a few important things. Thanks Jason!