path: root/doc/todo/tracking_bugs_with_dependencies.mdwn

[[!tag patch patch/core]]

I like the idea of [[tips/integrated_issue_tracking_with_ikiwiki]], and I do so on several wikis.  However, as far as I can tell, ikiwiki has no functionality which can represent dependencies between bugs and allow pagespecs to select based on dependencies.  For instance, I can't write a pagespec which selects all bugs with no dependencies on bugs not marked as done.  --[[JoshTriplett]]

> I started having a think about this.  I'm going to start with the idea that expanding
> the pagespec syntax is the way to attack this.  It seems that any pagespec that is going
> to represent "all bugs with no dependencies on bugs not marked as done" is going to
> need some way to represent "bugs not marked as done" as a collection of pages, and
> then represent "bugs which do not link to pages in the previous collection".
>
> One way to do this would be to introduce variables into the pagespec, along with
> universal and/or existential [[!wikipedia Quantification]].  That looks quite complex.
>
>> I thought about this briefly, and got about that far.. glad you got
>> further. :-) --[[Joey]]

>> Or, one [[!taglink could_also_refer|pagespec_in_DL_style]] to the language of [[!wikipedia description logics]]: their formulas actually define classes of objects through quantified relations to other classes. --Ivan Z.
> 
> Another option would be go with a more functional syntax.  The concept here would
> be to allow a pagespec to appear in a 'pagespec function' anywhere a page can.  e.g.
> I could pass a pagespec to `link()` and that would return true if there is a link to any
> page matching the pagespec.  This makes the variables and existential quantification
> implicit.  It would allow the example requested above:
>
>> `bugs/* and !*/Discussion and !link(bugs/* and !*/Discussion and !link(done))`
>
> Unfortunately, this is also going to make the pagespec parsing more complex because
> we now need to parse nested sets of parentheses to know when the nested pagespec
> ends, and that isn't a regular language (we can't use regular expression matching for
> easy parsing).
>
>> Also, it may cause ambiguities with page names that contain parens
>> (though some such ambigutities already exist with the pagespec syntax).
>
> One simplification of that would be to introduce some pagespec [[shortcuts]].  We could
> then allow pagespec functions to take either pages, or named pagespec shortcuts.  The
> pagespec shortcuts would just be listed on a special page, like current [[shortcuts]].
> (It would probably be a good idea to require that shortcuts on that page can only refer
> to named pagespecs higher up that page than themselves.  That would stop some
> looping issues...)  These shortcuts would be used as follows: when trying to match
> a page (without globs) you look to see if the page exists.  If it does then you have a
> match.  If it doesn't, then you look to see if a similarly named pagespec shortcut
> exists.  If it does, then you check that pagespec recursively to see if you have a match.
> The ordering requirement on named pagespecs stops infinite recursion.
>
> Does that seem like a reasonable first approach?
>
> -- [[Will]]

>> Having a separate page for the shortcuts feels unwieldly.. perhaps
>> instead the shortcut could be defined earlier in the scope of the same
>> pagespec that uses it?
>> 
>> Example: `define(~bugs, bugs/* and !*/Discussion) and define(~openbugs, ~bugs and !link(done)) and ~openbugs and !link(~openbugs)`

>>> That could work.  parens are only ever nested 1 deep in that grammar so it is regular and the current parsing would be ok.

>> Note that I made the "~" explicit, not implicit, so it could be left out. In the case of ambiguity between
>> a definition and a page name, the definition would win.

>>> That was my initial thought too :), but when implementing it I decided that requiring the ~ made things easier.  I'll probably require the ~ for the first pass at least.

>> So, equivilant example: `define(bugs, bugs/* and !*/Discussion) and define(openbugs, bugs and !link(done)) and openbugs and !link(openbugs)`
>> 

>> Re recursion, it is avoided.. but building a pagespec that is O(N^X) where N is the
>> number of pages in the wiki is not avoided. Probably need to add DOS prevention.
>>  --[[Joey]]

>>> If you memoize the outcomes of the named pagespecs you can make in O(N.X), no?
>>> -- [[Will]]

>>>> Yeah, guess that'd work. :-)

> <a id="another_kind_of_links" />One quick further thought.  All the above discussion assumes that 'dependency' is the
> same as 'links to', which is not really true.  For example, you'd like to be able to say
> "This bug does not depend upon [ [ link to other bug ] ]" and not have a dependency.
> Without having different types of links, I don't see how this would be possible.
>
> -- [[Will]]

>> I saw that this issue is targeted at by the work on [[structured page data#another_kind_of_links]]. --Ivan Z.

Okie - I've had a quick attempt at this.  Initial patch attached.  This one doesn't quite work.
And there is still a lot of debugging stuff in there.

At the moment I've added a new preprocessor plugin, `definepagespec`, which is like
shortcut for pagespecs.  To reference a named pagespec, use `~` like this:

    [ [!definepagespec name="bugs" spec="bugs/* and !*/Discussion"]]
    [ [!definepagespec name="openbugs" spec="~bugs and !link(done)"]]
    [ [!definepagespec name="readybugs" spec="~openbugs and !link(~openbugs)"]]

At the moment the problem is in `match_link()` when we're trying to find a sub-page that
matches the appropriate page spec.  There is no good list of pages available to iterate over.

    foreach my $nextpage (keys %IkiWiki::pagesources)

does not give me a good list of pages.  I found the same thing when I was working on
this todo [[todo/Add_a_plugin_to_list_available_pre-processor_commands]].

> I'm not sure why iterating over `%pagesources` wouldn't work here, it's the same method
> used by anything that needs to match a pagespec against all pages..? --[[Joey]]

>> My uchecked hypothesis is that %pagesources is created after the refresh hook.
>> I've also been concerned about how globally defined pagespec shortcuts would interact with
>> the page dependancy system.  Your idea of internally defined shortcuts should fix that. -- [[Will]]

>>> You're correct, the refresh hook is run very early, before pagesources
>>> is populated. (It will be partially populated on a refresh, but will
>>> not be updated to reflect new pages.) Agree that internally defined
>>> seems the way to go. --[[Joey]]

Immediately below is a patch which seems to basically work.  Lots of debugging code is still there
and it needs a cleanup, but I thought it worth posting at this point.  (I was having problems
with old style glob lists, so i just switched them off for the moment.)

The following three inlines work for me with this patch:

    Bugs:
    
    [ [!inline pages="define(~bugs, bugs/* and ! */Discussion) and ~bugs" archive="yes"]]
    
    OpenBugs:
    
    [ [!inline pages="define(~bugs, bugs/* and ! */Discussion) and define(~openbugs,~bugs and !link(done)) and ~openbugs" archive="yes"]]
    
    ReadyBugs:
    
    [ [!inline pages="define(~bugs, bugs/* and ! */Discussion) and define(~openbugs,~bugs and !link(done)) and define(~readybugs,~openbugs and !link(~openbugs)) and ~readybugs" archive="yes"]]

> Nice! Could the specfuncsref be passed in %params? I'd like to avoid
> needing to change the prototype of every pagespec function, since several
> plugins define them too. --[[Joey]]

>> Maybe - it needs more thought.  I also considered it when I was going though changing all those plugins :).
>> My concern was that `%params` can contain other user-defined parameters,
>> e.g. `link(target, otherparameter)`, and that means that the specFuncs could be clobbered by a user (or other
>> weird security hole).  I thought it better to separate it, but I didn't think about it too hard.  I might move it to
>> the first parameter rather than the second.  Ikiwiki is my first real perl hacking and I'm still discovering
>> good ways to write things in perl.
>>
>>>> `%params` contains the parameters passed to `pagespec_match`, not
>>>> user-supplied parameters. The user-supplied parameter to a function
>>>> like `match_glob()` or `match_link()` is passed in the second positional parameter. --[[Joey]]

>>>>> OK.  That seems reasonable then.  The only problem is that my PERLfu is not strong enough to make it
>>>>> work.  I really have to wonder what substance was influencing the designers of PERL...
>>>>> I can't figure out how to use the %params.  And I'm pissed off enough with PERL that I'm not going
>>>>> to try and figure it out any more.  There are two patches below now.  The first one uses an extra
>>>>> argument and works.  The second one tries to use %params and doesn't - take your pick :-). -- [[Will]]

>> What do you think is best to do about `is_globlist()`?  At the moment it requires that the 'second word', as
>> delimited by a space and ignoring parens, is 'and' or 'or'.  This doesn't hold in the above example pagespecs (so I just hard wired it to 0 to test my patch).
>> My thought was just to search for 'and' or 'or' as words anywhere in the pagespec.  Thoughts?

>>> Dunno, we could just finish deprecating it. Or change the regexp to
>>> skip over spaces in parens. (`/[^\s]+\s+([^)]+)/`) --[[Joey]]

>>>> I think I have a working regexp now.

>> Oh, one more thing.  In pagespec_translate (now pagespec_makeperl), there is a part of the regular expression for `# any other text`.
>> This contained `()`, which has no effect.  I replaced that with `\(\)`, but that is a change in the definition of pagespecs unrelated to the
>> rest of this patch. In a related change, commands were not able to contain `)` in their parameters.  I've extended that so the cannot
>> contain `(` or `)`.  -- [[Will]]

>>> `[^\s()]+` is a character class matching all characters not spaces or
>>> parens. Since the pervious terminals in the regexp consume most
>>> occurances of an open paren or close paren, it's unlikely for one to
>>> get through to that part of the regexp. For example, "foo()" will be
>>> matched by the command matcher; "(foo)" will be matched by the open
>>> paren literal terminal. "foo(" and "foo)" can get through to the
>>> end, and would be matched as a page name, if it didn't exclude parens.
>>>
>>> So why exclude them? Well, consider "foo and(bar and baz)". We don't
>>> want it to match "and(" as a page name!
>>> 
>>> Escaping the parens in the character class actually changes nothing; the
>>> changed character class still matches all characters not spaces or
>>> parens. (Try it!).
>>> 
>>> Re commands containing '(', I don't really see any reason not to
>>> allow that, unless it breaks something. --[[Joey]]

>>>> Oh, I didn't realise you didn't need to escape parens inside [].  All else I
>>>> I understood.  I have stopped commands from containing parens because
>>>> once you allow that then you might have a extra level of depth in the parsing
>>>> of define() statements. -- [[Will]]

>>> Updated patch.  Moved the specFuncsRef to the front of the arg list.  Still haven't thought through the security implications of
>>> having it in `%params`.  I've also removed all the debugging `print` statements.  And I've updated the `is_globlist()` function.
>>> I think this is ready for people other than me to have a play.  It is not well enough tested to commit just yet.
>>> -- [[Will]]

I've lost track of the indent level, so I'm going back to not indented - I think this is a working [[patch]] taking into
account all comments above (which doesn't mean it is above reproach :) ).  --[[Will]]

> Very belated code review of last version of the patch:
> 
> * `is_globlist` is no longer needed

>> Good :)

> * I don't understand why the pagespec match regexp is changed
>   from having flags `igx` to `ixgs`. Don't see why you
>   want `.` to match '\n` in it, and don't see any `.` in the regexp 
>   anyway?

>> Because you have to define all the named pagespecs in the pagespec, you sometimes end up with very long pagespecs.  I found it useful to split them over multiple lines.  That didn't work at one point and I added the 's' to make it work.  I may have further altered the regex since then to make the 's' redundant.  Remove it and see if multi-line pagespecs still work. :)

>>> Well, I can tell you that multi-line pagespecs are supported w/o
>>> your patch .. I use them all the time. The reason I find your
>>> use of `/s` unlikely is because without it `\s` already matches
>>> a newline. Only if you want to treat a newline as non-whitespace
>>> is `/s` typically necessary. --[[Joey]] 

> * Some changes of `@_` to `%params` in `pagespec_makeperl` do not
>   make sense to me. I don't see where \%params is defined and populated,
>   except with `\$params{specFunc}`.

>> I'm not a perl hacker.  This was a mighty battle for me to get going.
>> There is probably some battlefield carnage from my early struggles
>> learning perl left here. Part of this is that @_ / @params already
>> existed as a way of passing in extra parameters.  I didn't want to
>> pollute that top level namespace - just at my own parameter (a hash)
>> which contained the data I needed.

>>> I think I understand how the various `%params`
>>> (there's not just one) work in your code now, but it's really a mess.
>>> Explaining it in words would take pages.. It could be fixed by,
>>> in `pagespec_makeperl` something like:
>>> 
>>>	my %specFuncs;
>>>	push @_, specFuncs => \%specFuncs;
>>> 
>>> With that you have the hash locally available for populating
>>> inside `pagespec_makeperl`, and when the `match_*` functions
>>> are called the same hash data will be available inside their
>>> `@_` or `%params`. No need to change how the functions are called
>>> or do any of the other hacks.
>>>
>>> Currently, specFuncs is populated by building up code
>>> that recursively calls `pagespec_makeperl`, and is then
>>> evaluated when the pagespec gets evaluated. My suggested
>>> change to `%params` will break that, but that had to change 
>>> anyway.
>>>
>>> It probably has a security hole, and is certianly inviting
>>> one, since the pagespec definition is matched by a loose regexp (`.*`)
>>> and then subject to string interpolation before being evaluated
>>> inside perl code. I recently changed ikiwiki to never interpolate
>>> user-supplied strings when translating pagespecs, and that
>>> needs to happen here too. The obvious way, it seems to me,
>>> is to not generate perl code, but just directly run perl code that
>>> populates specFuncs.

>>>> I don't think this is as bad as you make out, but your addition of the
>>>> data array will break with the recursion my patch adds in pagespec_makeperl.
>>>> To fix that I'll need to pass a reference to that array into pagespec_makeperl.
>>>> I think I can then do the same thing to $params{specFuncs}.  -- [[Will]]

>>>>> You're right -- I did not think the recursive case through.
>>>>> --[[Joey]] 

> * Seems that the only reason `match_glob` has to check for `~` is
>   because when a named spec appears in a pagespec, it is translated
>   to `match_glob("~foo")`. If, instead, `pagespec_makeperl` checked
>   for named specs, it could convert them into `check_named_spec("foo")`
>   and avoid that ugliness.

>> Yeah - I wanted to make named specs syntactically different on my first pass.  You are right in that this could be made a fallback - named specs always override pagenames.

> * The changes to `match_link` seem either unecessary, or incomplete.
>   Shouldn't it check for named specs and call
>   `check_named_spec_existential`?

>>  An earlier version did.  Then I realised it wasn't actually needed in that case - match_link() already included a loop that was like a type of existential matching.  Each time through the loop it would
>> call match_glob().  match_glob() in turn will handle the named spec.  I tested this version briefly and it seemed to work.  I remember looking at this again later and wondering if I had mis-understood
>> some of the logic in match_link(), which might mean there are cases where you would need an explicit call to check_named_spec_existential() - I never checked it properly after having that thought.

>>> In the common case, `match_link` does not call `match_glob`,
>>> because the link target it is being asked to check for is a single
>>> page name, not a glob.

>>>> A named pagespec should fall into the glob case.  These two pagespecs should be the same:

    link(a*)

>>>> and

    define(aStar, a*) and link(~aStar)

>>>> In the first case, we want the pagespec to match any page that links to a page matching the glob.
>>>> In the second case, we want the pagespec to match any page that links to a page matching the named spec.
>>>> match_link() was already doing existential part.  The patches to this code were simply to remove the `lc()`
>>>> call from the named pagespec name.  Can that `lc` be removed entirely? -- [[Will]]

>>>>> I think we could get rid of it. `bestlink` will lc it itself
>>>>> if the uppercase version does not exist; `match_glob` matches
>>>>> insensitively.
>>>>> --[[Joey]] 

> * Generally, the need to modify `match_*` functions so that they
>   check for and handle named pagespecs seems suboptimal, if
>   only because there might be others people may want to use named
>   pagespecs with. It would be possible to move this check
>   to `pagespec_makeperl`, by having it check if the parameter
>   passed to a pagespec function looked like a named pagespec.
>   The only issue is that some pagespec functions take a parameter
>   that is not a page name at all, and it could be weird
>   if such a parameter were accidentially interpreted as a named
>   pagespec. (But, that seems unlikely to happen.)

>> Possibly.  I'm not sure which I prefer between the current solution and that one.  Each have advantages and disadvantages.
>> It really isn't much code for the match functions to add a call to check_named_spec_existential().

>>> But if a plugin adds its own match function, it has
>>> to explicitly call that code to support named pagespecs.

>>>> Yes, and it can do that in just three lines of code.  But if we automatically check for named pagespecs all the time we
>>>> potentially break any matching function that doesn't accept pages, or wants to use multiple arguments.

>>>>> 3 lines of code, plus the functions called become part of the API,
>>>>> don't forget about that..
>>>>>
>>>>> Yes, I think that is the tradeoff, the question is whether to export
>>>>> the additional complexity needed for that flexability.
>>>>>
>>>>> I'd be suprised if multiple argument pagespecs become necessary..
>>>>> with the exception of this patch there has been no need for them yet.
>>>>>
>>>>> There are lots of pagespecs that take data other than pages,
>>>>> indeed, that's really the common case. So far, none of them 
>>>>> seem likely to take data that starts with a `~`. Perhaps
>>>>> the thing to do would be to check if `~foo` is a known,
>>>>> named pagespec, and if not, just pass it through unchanged.
>>>>> Then there's little room for ambiguity, and this also allows
>>>>> pagespecs like `glob(~foo*)` to match the literal page `~foo`.
>>>>> (It will make pagespec_merge even harder tho.. see below.)
>>>>> --[[Joey]] 

>>>>>> I've already used multi-argument pagespec match functions in
>>>>>> my data plugin.  It is used for having different types of links.  If
>>>>>> you want to have multiple types of links, then the match function
>>>>>> for them needs to take both the link name and the link type.
>>>>>> I'm trying to think of a way we could have both - automatically
>>>>>> handle the existential case unless the function indicates somehow
>>>>>> that it'll do it itself.  Any ideas?  -- [[Will]]

> * I need to check if your trick to avoid infinite recursion
>   works if there are two named specs that recursively
>   call one-another. I suspect it does, but will test this
>   myself..

>> It worked for me. :)

> * I also need to verify if memoizing the named pagespecs has
>   really guarded against very expensive pagespecs DOSing the wiki..

> --[[Joey]] 

>>  There is one issue that I've been thinking about that I haven't raised anywhere (or checked myself), and that is how this all interacts with page dependencies.
>>  Firstly, I'm not sure anymore that the `pagespec_merge` function will continue to work in all cases.

>>> The problem I can see there is that if two pagespecs
>>> get merged and both use `~foo` but define it differently,
>>> then the second definition might be used at a point when
>>> it shouldn't (but I haven't verified that really happens).
>>> That could certianly be a show-stopper. --[[Joey]] 

>>>> I think this can happen in the new closure based code.  I don't think this could happen in the old code.  -- [[Will]]

>>>> Even if that works, this is a good argument for having a syntactic difference between named pagespecs and normal pages.
>>>> If you're joining two pagespecs with 'or', you don't want a named pagespec in the first part overriding a page name in the
>>>> second part.  Oh, and I assume 'or' has the right operator precedence that "a and b or c" is "(a and b) or c", and not "a and (b or c)" -- [[Will]]

>>>>> Looks like its bracketed in the code anyway... -- [[Will]]

>>>> Perhaps the thing to do is to have a `clear_defines()`
>>>> function, then merging `A` and `B` yields `(A) or (clear_defines() and (B))`
>>>> That would deal with both the cases where `A` and `B` differently
>>>> define `~foo` as well as with the case where `A` defines `~foo` while
>>>> `B` uses it to refer to a literal page.
>>>> --[[Joey]]

>>>>> I don't think this will work with the new patch, and I don't think it was needed with the old one.
>>>>> Under the old patch, pagespec_makeperl() generated a string of unevaluated, self-contained, perl
>>>>> code.  When a new named pagespec was defined, a recursive call was made to get the perl code
>>>>> for the pagespec, and then that code was used to add something like `$params{specFuncs}->{name} = sub {recursive code} and `
>>>>> to the result of the calling function.  This means that at pagespec testing time, when this code is executed, the
>>>>> specFuncs hash is built up as the pagespec is checked.  In the case of the 'or' used above, later redefinitions of
>>>>> a named pagespec would have redefined the specFunc at the right time.  It should have just worked.  However...

>>>>> Since my original patch, you started using closures for security reasons (and I can see the case for that).  Unfortunately this
>>>>> means that the generated perl code is no longer self-contained - it needs to be evaluated in the same closure it was generated
>>>>> so that it has access to the data array.  To make this work with the recursive call I had two options: a) make the data array a
>>>>> reference that I pass around through the pagespec_makeperl() functions and have available when the code is finally evaluated
>>>>> in pagespec_translate(), or b) make sure that each pagespec is evaluated in its correct closure and a perl function is returned, not a
>>>>> string containing unevaluated perl code.

>>>>> I went with option b).  I did it in such a way that the hash of specfuncs is built up at translation time, not at execution time.  This
>>>>> means that with the new code you can call specfuncs that get defined out of order:

    ~test and define(~test, blah)

>>>>> but it also means that using a simple 'or' to join two pagespecs wont work.  If you do something like this:

    ~test and define(~test, foo) and define(~test, baz)

>>>>> then the last definition (baz) takes precedence.
>>>>> In the process of writing this I think I've come up with a way to change this back the way it was, still using closures. -- [[Will]]

>>> Alternatively, my [[remove-pagespec-merge|should_optimise_pagespecs]]
>>> branch solves this, in a Gordian knot sort of way :-) --[[smcv]]

>>  Secondly, it seems that there are two types of dependency, and ikiwiki
>>  currently only handles one of them.  The first type is "Rebuild this
>>  page when any of these other pages changes" - ikiwiki handles this.
>>  The second type is "rebuild this page when set of pages referred to by
>>  this pagespec changes" - ikiwiki doesn't seem to handle this.  I
>>  suspect that named pagespecs would make that second type of dependency
>>  more important.  I'll try to come up with a good example. -- [[Will]]

>>> Hrm, I was going to build an example of this with backlinks, but it
>>> looks like that is handled as a special case at the moment (line 458 of
>>> render.pm).  I'll see if I can breapk
>>> things another way.  Fixing this properly would allow removal of that special case. -- [[Will]]

>>>> I can't quite understand the distinction you're trying to draw
>>>> between the two types of dependencies. Backlinks are a very special
>>>> case though and I'll be suprised if they fit well into pagespecs.
>>>> --[[Joey]] 

>>>>> The issue is that the existential pagespec matching allows you to build things that have similar
>>>>> problems to backlinks.
>>>>> e.g. the following inline:

    \[[!inline pages="define(~done, link(done)) and link(~done)" archive=yes]]

>>>>> includes any page that links to a page that links to done.  Now imagine I add a new link to 'done' on
>>>>> some random page somewhere - a page which some other page links to which didn't previously get included - the set of pages accepted by the pagespec, and hence the set of
>>>>> pages inlined, will change.  But, there is no dependency anywhere on the page that I altered, so
>>>>> ikiwiki will not rebuild the page with the inline in it.  What is happening is that the page that I altered affects
>>>>> the set of pages matched by the pagespec without itself being matched by the pagespec, and hence included in the dependency list.

>>>>> To make this work well, I think you need to recognise two types of dependencies for each page (and no
>>>>> special cases for particular types of links, eg backlinks).  The first type of dependency says, "The content of
>>>>> this page depends upon the content of these other pages".  The `add_depends()` in the shortcuts
>>>>> plugin is of this form: any time the shortcuts page is edited, any page with a shortcut on it
>>>>> is rebuilt.  The inline plugin also needs to add dependencies of this form to detect when the inlined
>>>>> content changes.  By contrast, the map plugin does not need a dependency of this form, because it
>>>>> doesn't actually care about the content of any pages, just which pages it needs to include (which we'll handle next).

>>>>> The second type of dependency says, "The content of this page depends upon the exact set of pages matched
>>>>> by this pagespec".  The first type of dependency was about the content of some pages, the second type is about
>>>>> which pages get matched by a pagespec.  This is the type of dependency tracking that the map plugin needs.
>>>>> If the set of pages matched by map pagespec changes, then the page with the map on it needs to be rebuilt to show a different list of pages.
>>>>> Inline needs this type of dependency as well as the previous type - This type handles a change in which pages
>>>>> are inlined, the previous type handles a change in the content of any of those pages.  Shortcut does not need this type of
>>>>> dependency.  Most of the places that use `add_depends()` seem to need this type of dependency rather than the first type.

>>>>>> Note that inline and map currently achieve the second type of dependency by
>>>>>> explicitly calling `add_depends` for each page the displayed.
>>>>>> If any of those pages are removed, the regular pagespec would not
>>>>>> match them -- since they're gone. However, the explicit dependency
>>>>>> on them does cause them to match. It's an ugly corner I'd like to
>>>>>> get rid of. --[[Joey]]

>>>>> Implementation Details:  The first type of dependency can be handled very similarly to the current
>>>>> dependency system.  You just need to keep a list of pages that the content depends upon.  You could
>>>>> keep that list as a pagespec, but if you do this you might want to check that the pagespec doesn't change,
>>>>> possibly by adding a dependency of the second type along with the dependency of the first type.

>>>>>> An example of the current system not tracking enough data is 
>>>>>> where A inlines B which inlines C. A change to C will cause B to
>>>>>> rebuild, but A will not "notice" that B has implicitly changed.
>>>>>> That example suggests it might be fixable without explicitly storing
>>>>>> data, by causing a rebuild of B to be treated as a change to B.
>>>>>> --[[Joey]] 

>>>>> The second type of dependency is a little more tricky.  For each page, we'd need a list of pagespecs that
>>>>> the page depended on, and for each pagespec you'd want to store the list of pages that currently match it.
>>>>> On refresh, you'd need to check each pagespec to see if the set of pages that match it has changed, and if
>>>>> that set has changed, then rebuild the dependent page(s).  Oh, and for this second type of dependency, I
>>>>> don't think you can merge pagespecs.  If I wanted to know if either "\*" or "link(done)" changes, then just checking
>>>>> to see if the set of pages matched by "\* or link(done)" changes doesn't work.

>>>>> The current system works because even though you usually want dependencies of the second type, the set of pages
>>>>> referred to by a pagespec can only change if one of those pages itself changes.  i.e. A dependency check of the
>>>>> first type will catch a dependency change of the second type with current pagespecs.
>>>>> This doesn't work with backlinks, and it doesn't work with existential matching.  Backlinks are currently special-cased.  I don't know
>>>>> how to special-case existential matching - I suspect you're better off just getting the dependency tracking right.

>>>>> I also tried to come up with other possible solutions: e.g. can we find the dependencies for a pagespec?  That
>>>>> would be the set of pages where a change on one of those pages could lead to a change in the set of pages matched by the pagespec.
>>>>> For old-style pagespecs without backlinks, the dependency set for a pagespec is the same as the set of pages the pagespec matches.
>>>>> Unfortunately, with existential matching, the set of pages that each
>>>>> pagespec depends upon can quickly become "*", which is not very useful.  -- [[Will]]

Patch updated to use closures rather than inline generated code for named pagespecs.  Also includes some new use of ErrorReason where appropriate. -- [[Will]]

> * Perl really doesn't need forward declarations, honest!

>> It complained (warning, not error) when I didn't use the forward declaration. :(

> * I have doubts about memoizing the anonymous sub created by
>   `pagespec_translate`.

>> This is there explicitly to make sure that runtime is polynomial and not exponential.

> * Think where you wrote `+{}` you can just write `{}`

>> Possibly :) -- [[Will]]

----

    diff --git a/IkiWiki.pm b/IkiWiki.pm
    index 061a1c6..1e78a63 100644
    --- a/IkiWiki.pm
    +++ b/IkiWiki.pm
    @@ -1774,8 +1774,12 @@ sub pagespec_merge ($$) {
     	return "($a) or ($b)";
     }
     
    -sub pagespec_translate ($) {
    +# is perl really so dumb it requires a forward declaration for recursive calls?
    +sub pagespec_translate ($$);
    +
    +sub pagespec_translate ($$) {
     	my $spec=shift;
    +	my $specFuncsRef=shift;
     
     	# Convert spec to perl code.
     	my $code="";
    @@ -1789,7 +1793,9 @@ sub pagespec_translate ($) {
     		|
     			\)		# )
     		|
    -			\w+\([^\)]*\)	# command(params)
    +			define\(\s*~\w+\s*,((\([^()]*\)) | ([^()]+))+\) # define(~specName, spec) - spec can contain parens 1 deep
    +		|
    +			\w+\([^()]*\)	# command(params) - params cannot contain parens
     		|
     			[^\s()]+	# any other text
     		)
    @@ -1805,10 +1811,19 @@ sub pagespec_translate ($) {
     		elsif ($word eq "(" || $word eq ")" || $word eq "!") {
     			$code.=' '.$word;
     		}
    -		elsif ($word =~ /^(\w+)\((.*)\)$/) {
    +		elsif ($word =~ /^define\(\s*(~\w+)\s*,(.*)\)$/s) {
    +			my $name = $1;
    +			my $subSpec = $2;
    +			my $newSpecFunc = pagespec_translate($subSpec, $specFuncsRef);
    +			return if $@ || ! defined $newSpecFunc;
    +			$specFuncsRef->{$name} = $newSpecFunc;
    +			push @data, qq{Created named pagespec "$name"};
    +			$code.="IkiWiki::SuccessReason->new(\$data[$#data])";
    +		}
    +		elsif ($word =~ /^(\w+)\((.*)\)$/s) {
     			if (exists $IkiWiki::PageSpec::{"match_$1"}) {
     				push @data, $2;
    -				$code.="IkiWiki::PageSpec::match_$1(\$page, \$data[$#data], \@_)";
    +				$code.="IkiWiki::PageSpec::match_$1(\$page, \$data[$#data], \@_, specFuncs => \$specFuncsRef)";
     			}
     			else {
     				push @data, qq{unknown function in pagespec "$word"};
    @@ -1817,7 +1832,7 @@ sub pagespec_translate ($) {
     		}
     		else {
     			push @data, $word;
    -			$code.=" IkiWiki::PageSpec::match_glob(\$page, \$data[$#data], \@_)";
    +			$code.=" IkiWiki::PageSpec::match_glob(\$page, \$data[$#data], \@_, specFuncs => \$specFuncsRef)";
     		}
     	}
     
    @@ -1826,7 +1841,7 @@ sub pagespec_translate ($) {
     	}
     
     	no warnings;
    -	return eval 'sub { my $page=shift; '.$code.' }';
    +	return eval 'memoize (sub { my $page=shift; '.$code.' })';
     }
     
     sub pagespec_match ($$;@) {
    @@ -1839,7 +1854,7 @@ sub pagespec_match ($$;@) {
     		unshift @params, 'location';
     	}
     
    -	my $sub=pagespec_translate($spec);
    +	my $sub=pagespec_translate($spec, +{});
     	return IkiWiki::ErrorReason->new("syntax error in pagespec \"$spec\"")
     		if $@ || ! defined $sub;
     	return $sub->($page, @params);
    @@ -1850,7 +1865,7 @@ sub pagespec_match_list ($$;@) {
     	my $spec=shift;
     	my @params=@_;
     
    -	my $sub=pagespec_translate($spec);
    +	my $sub=pagespec_translate($spec, +{});
     	error "syntax error in pagespec \"$spec\""
     		if $@ || ! defined $sub;
     	
    @@ -1872,7 +1887,7 @@ sub pagespec_match_list ($$;@) {
     sub pagespec_valid ($) {
     	my $spec=shift;
     
    -	my $sub=pagespec_translate($spec);
    +	my $sub=pagespec_translate($spec, +{});
     	return ! $@;
     }
     
    @@ -1919,6 +1934,68 @@ sub new {
     
     package IkiWiki::PageSpec;
     
    +sub check_named_spec($$;@) {
    +	my $page=shift;
    +	my $specName=shift;
    +	my %params=@_;
    +
    +	return IkiWiki::ErrorReason->new("Unable to find specFuncs in params to check_named_spec()!")
    +		unless exists $params{specFuncs};
    +
    +	my $specFuncsRef=$params{specFuncs};
    +
    +	return IkiWiki::ErrorReason->new("Named page spec '$specName' is not valid")
    +		unless (substr($specName, 0, 1) eq '~');
    +
    +	if (exists $specFuncsRef->{$specName}) {
    +		# remove the named spec from the spec refs
    +		# when we recurse to avoid infinite recursion
    +		my $sub = $specFuncsRef->{$specName};
    +		delete $specFuncsRef->{$specName};
    +		my $result = $sub->($page, %params);
    +		$specFuncsRef->{$specName} = $sub;
    +		return $result;
    +	} else {
    +		return IkiWiki::ErrorReason->new("Page spec '$specName' does not exist");
    +	}
    +}
    +
    +sub check_named_spec_existential($$$;@) {
    +	my $page=shift;
    +	my $specName=shift;
    +	my $funcref=shift;
    +	my %params=@_;
    +
    +	return IkiWiki::ErrorReason->new("Unable to find specFuncs in params to check_named_spec_existential()!")
    +			unless exists $params{specFuncs};
    +	my $specFuncsRef=$params{specFuncs};
    +	
    +	return IkiWiki::ErrorReason->new("Named page spec '$specName' is not valid")
    +		unless (substr($specName, 0, 1) eq '~');
    +
    +	if (exists $specFuncsRef->{$specName}) {
    +		# remove the named spec from the spec refs
    +		# when we recurse to avoid infinite recursion
    +		my $sub = $specFuncsRef->{$specName};
    +		delete $specFuncsRef->{$specName};
    +		
    +		foreach my $nextpage (keys %IkiWiki::pagesources) {
    +			if ($sub->($nextpage, %params)) {
    +				my $tempResult = $funcref->($page, $nextpage, %params);
    +				if ($tempResult) {
    +					$specFuncsRef->{$specName} = $sub;
    +					return IkiWiki::SuccessReason->new("Existential check of '$specName' matches because $tempResult");
    +				}
    +			}
    +		}
    +		
    +		$specFuncsRef->{$specName} = $sub;
    +		return IkiWiki::FailReason->new("No page in spec '$specName' was successfully matched");
    +	} else {
    +		return IkiWiki::ErrorReason->new("Named page spec '$specName' does not exist");
    +	}
    +}
    +
     sub derel ($$) {
     	my $path=shift;
     	my $from=shift;
    @@ -1937,6 +2014,10 @@ sub match_glob ($$;@) {
     	my $glob=shift;
     	my %params=@_;
     	
    +	if (substr($glob, 0, 1) eq '~') {
    +		return check_named_spec($page, $glob, %params);
    +	}
    +
     	$glob=derel($glob, $params{location});
     
     	my $regexp=IkiWiki::glob2re($glob);
    @@ -1959,8 +2040,9 @@ sub match_internal ($$;@) {
     
     sub match_link ($$;@) {
     	my $page=shift;
    -	my $link=lc(shift);
    +	my $fullLink=shift;
     	my %params=@_;
    +	my $link=lc($fullLink);
     
     	$link=derel($link, $params{location});
     	my $from=exists $params{location} ? $params{location} : '';
    @@ -1975,25 +2057,37 @@ sub match_link ($$;@) {
     		}
     		else {
     			return IkiWiki::SuccessReason->new("$page links to page $p matching $link")
    -				if match_glob($p, $link, %params);
    +				if match_glob($p, $fullLink, %params);
     			$p=~s/^\///;
     			$link=~s/^\///;
     			return IkiWiki::SuccessReason->new("$page links to page $p matching $link")
    -				if match_glob($p, $link, %params);
    +				if match_glob($p, $fullLink, %params);
     		}
     	}
     	return IkiWiki::FailReason->new("$page does not link to $link");
     }
     
     sub match_backlink ($$;@) {
    -	return match_link($_[1], $_[0], @_);
    +	my $page=shift;
    +	my $backlink=shift;
    +	my @params=@_;
    +
    +	if (substr($backlink, 0, 1) eq '~') {
    +		return check_named_spec_existential($page, $backlink, \&match_backlink, @params);
    +	}
    +
    +	return match_link($backlink, $page, @params);
     }
     
     sub match_created_before ($$;@) {
     	my $page=shift;
     	my $testpage=shift;
     	my %params=@_;
    -	
    +
    +	if (substr($testpage, 0, 1) eq '~') {
    +		return check_named_spec_existential($page, $testpage, \&match_created_before, %params);
    +	}
    +
     	$testpage=derel($testpage, $params{location});
     
     	if (exists $IkiWiki::pagectime{$testpage}) {
    @@ -2014,6 +2108,10 @@ sub match_created_after ($$;@) {
     	my $testpage=shift;
     	my %params=@_;
     	
    +	if (substr($testpage, 0, 1) eq '~') {
    +		return check_named_spec_existential($page, $testpage, \&match_created_after, %params);
    +	}
    +
     	$testpage=derel($testpage, $params{location});
     
     	if (exists $IkiWiki::pagectime{$testpage}) {