Wikipedia:Lua/Requests/Archive 3

I was just looking into this on my own. It is definitely possible to put {{REVISIONTIMESTAMP}} on a page (for example, in a template, flanked by includeonly tags) and to access it using frame:expandTemplate [7]. However, it is a serious pain in the ass that REVISIONTIMESTAMP:Some_page doesn't work. So far I have not managed to figure out a Lua function to work with it either, even without parameters. There are a LOT of variations there, though, and I may not have tested the right one at all. Can someone confirm if it's impossible? Even as it is -- we could tag all the relevant articles on the article page with a template that contains REVISIONTIMESTAMP, and read them all in, and sort it out using some delimiter that marks the template. It would just be really ugly coding. Wnt (talk) 20:10, 5 November 2013 (UTC)

{{REVISIONTIMESTAMP:Some_page}} is T8092, which interestingly enough was just fixed a day or two ago. The change should be live here with 1.23wmf3, which is scheduled for deployment here on November 14. Anomie⚔ 02:03, 6 November 2013 (UTC)

Cool! What I propose (and will try to code then if I get the chance) is a simple Template:Linked revisions with a Lua innard, that takes a page name, date, and search template (usually a namespace) as parameters. It runs the full text of the named page through a simple filter that pulls out everything between a pair of double brackets, tries to interpret that as a page name, checks whether it fits the search template (i.e. does it match, say, "Talk:") if one is provided, looks up the REVISIONTIMESTAMP for it if so, and if that revisiontimestamp is after the date specified, appends the pagename, REVISIONUSER, and date last edited to the output if so. The result is that you can look at "all edits to any article linked from Bob Dole in the past two weeks, and who made them". (Admittedly, I suspect good and evil will benefit almost equally from this tool, but let's hope the angels edge forward by a nose) Wnt (talk) 05:31, 6 November 2013 (UTC)

Err, Special:RecentChangesLinked/Bob Dole already does basically this, and won't run into the expensive parser function limit. Why do we need a replacement in Lua? Anomie⚔ 11:17, 6 November 2013 (UTC)

Heh. A) because I forgot that feature existed, or at least, thought it was something I'd once seen on the toolserver, B) because the Lua feature can be programmed to process the output all sorts of different ways (and though I'm not sure, but I think the limit can be evaded by using the magic word for the lookups). For example, I think I should add an ending date, so that I can get a list of articles linked from a parent (say,

Cassini probe) that haven't been touched in over a year. Wnt (talk

) 15:30, 6 November 2013 (UTC)

I don not exactly understand this, but some notes. 1. Over at template talk:RELC list I am cooperating with a bot programmer to make this working for all projects (generic), and even wider (like tracking multipage templates). The bot will list all project pages (through the project talkpage template). Templates make the special page available for the user. 3. And the Request template (similar task as the archivebot-template) can have multiple options (eg exclude/include certain pages). This is to inform you, and maybe prevent double works. -DePiep (talk) 16:57, 7 November 2013 (UTC)

This should be complementary. The template will allow a module to sniff out the appropriate article list (like

Wikipedia:WikiProject Medicine/List of pages/Articles). The Lua module should read in that list (just one mw.title access). Then I go fish - get the REVISIONTIMESTAMP for each link on that page. There's nothing in the Lua reference manual indicating that frame:expandTemplate() is expensive (though perchance it may be by the time I finish abusing it) so I picture just running a template containing the magic word and its parameter. (There's still a chance that there's some other function that will let me do the magic word directly without making it a template, and I won't know if anything really works until I see it happen) After that I can start adorning the module with features to screen down the output (another few ideas: screen out edits from a list of known patrollers; list only edits by IPs). Wnt (talk

) 23:23, 7 November 2013 (UTC)

OK, {{REVISIONTIMESTAMP:Lua}} is producing 20210713202336, which is a good sign... Wnt (talk) 20:25, 14 November 2013 (UTC)

I've posted Module:Related changes as a start; I still need to write a way to read in a bunch of anonymous parameters and interpret them as filter options, track down where the expensive calls come in, etc., but it's already potentially useful. Wnt (talk) 07:31, 15 November 2013 (UTC)

Don't know if there is an overlap, but over at Page reports a botmaster could be walking in this field too. -DePiep (talk) 10:01, 15 November 2013 (UTC)

Unfortunately, it appears that every use of REVISIONUSER or REVISIONTIMESTAMP with a parameter is counted as an expensive call, so I think I should leave this one to the botmasters. It really isn't possible to patrol all the links from an article, let alone a project, with a limit of 500. Unless, that is, there's some way to connive at the function from Lua that avoids this whole infrastructure (I never have figured out all of those ways of calling parser functions from Lua, etc.) but I doubt it... Wnt (talk) 04:41, 17 November 2013 (UTC)

I have a question, I have made the edits to the Module:Authority control/sandbox to update it to get the NDL property on the wikidata page associated with wiki page, but for somereason it doesn't seem to work. Can someone take a look to see what is wrong? --Clarkcj12 (talk) 13:17, 17 November 2013 (UTC)

The key here is to try to get testcases working. I started at Module:Authority control/sandbox/testcases - however, the only way I can think of to get the answers so that the test can be set to actually succeed would be to reverse engineer from the HMTL source. Substing the template module doesn't get the wikicode. (for my confusion, see the history of the talk page for that) Plus, I don't think I'm actually using the sandbox code. Apart from that, there's the specific situation: I have no idea what the NDL property is and what a successful call with it would look like (That is Clarkcj12's responsibility, I think). But in general we need a much clearer documentation of how to start up and use sandbox testing in the first place, if we want anyone but admins ever to touch protected templates. Wnt (talk) 20:28, 17 November 2013 (UTC)

Wnt is correct that a set of proper testcases is needed. A quick look suggests the new code is fine, although the first step in editing a sandbox is to copy the existing main module (several things in the main module are not in the sandbox). What doesn't work? Bear in mind that people like me have only the vaguest idea about the topic—what wikitext posted where fails to produce the expected output, and what is the expected output? By the way, pairs( conf ) is used twice, whereas it's likely that ipairs( conf ) is wanted (the latter will consistently iterate the table in the order it is defined as the table has only numbered elements). Johnuniq (talk) 01:26, 18 November 2013 (UTC)

@Wnt: Special:ExpandTemplates can be used to capture the output of {{Authority control}}. I have fiddled with a test program I wrote as an alternative. It is for testing Module:Convert and has some problems when applied more generally, but I've put a few tests at User:Johnuniq/sandbox4. The table on that page is very wide—if you sroll to the right you can see "Pass" next to each test. Johnuniq (talk) 03:50, 18 November 2013 (UTC)

I had another look and have fixed the problem, I think. The new code in Module:Authority control/sandbox had a typo (function ndlLInk had an uppercase "I"), so I fixed that. First, I copied Module:Authority control to the sandbox, then I'm afraid I severely munged it to do two things: fix the whitespace for the new system of using tabs for indents; and insert "local" where desirable (which meant the functions then had to be re-ordered so they were defined in the correct order). Finally, I put the new NDL code into the sandbox. Here is a test (using the sandbox template which calls the sandbox module):

• {{Authority control/sandbox|NDL=01013687}}

The next step would be test the code and decide if any other changes should be made. Then have the editors who maintain the template/module consider the changes. Johnuniq (talk) 05:22, 18 November 2013 (UTC)

Well, I've updated the sandbox/testcases. It works... my, it's an ugly process though. Even using the module filed away under Module:Convert/tester, the user then needs to manually copy and paste a section because subst doesn't work, and make null edits each time so the error/out of date output is removed from page self processing. I wish we had a more user friendly way, because I think more than half the users coming here, we tell them, or ought to tell them, please, give us testcases! Wnt (talk) 07:57, 18 November 2013 (UTC)

Agreed, it's ugly. A better way to generate tests would be very helpful. By the way, rather than null edits, add ?action=purge to the URL to purge the page. That can be automated using {{purge}}, as shown in the first line of Module talk:Convert/sandbox/testcases. Johnuniq (talk) 08:45, 18 November 2013 (UTC)

Hmmm, it's been a while since I toyed with that. I remember early on experimenting and feeling like going a literal null edit was 'more effective' than purging somehow, but it's been so long I don't remember what circumstances that is in. Wnt (talk) 08:50, 18 November 2013 (UTC)

Task

Lua replacement of {{cite doi}} and {{cite pmid}} templates to allow passing of optional parameters to the underlying {{cite journal}} template.

Context

Currently the {{

• mandatory parameter:
  PMID or digital object identifier
  (doi)
• list of one or more optional parameters to be passed to the underlying {{cite journal}} template

Testcase

{{Cite pmid|12122621|authorformat = vanc | author-separator=, | author-name-separator = }}

Based on:

• Ozturan, O.; Miman, M. C.; Yiğit, B.; Cokkeser, Y.; Kizilay, A.; Aktaş, D. (2002). "Approaches to twisted noses and results of treatment". Kulak burun bogaz ihtisas dergisi : KBB = Journal of ear, nose, and throat. 9 (1): 21–29.
  PMID 12122621. {{cite journal}}: Cite has empty unknown parameters: |author-name-separator=, |author-separator=, |authorformat=, and |displayauthors= (help
  )

• Ozturan, O.; Miman, M. C.; Yiğit, B.; Cokkeser, Y.; Kizilay, A.; Aktaş, D. (2002). "Approaches to twisted noses and results of treatment". Kulak burun bogaz ihtisas dergisi : KBB = Journal of ear, nose, and throat. 9 (1): 21–29.
  PMID 12122621. {{cite journal}}: Unknown parameter |author-name-separator= ignored (help); Unknown parameter |author-separator= ignored (help); Unknown parameter |authorformat= ignored (help
  )

If something similar has been done before, I would appreciate a link. With a good example, I may be able to code this myself. Thanks. Boghog (talk) 15:51, 23 November 2013 (UTC)

Hi, I'm one of the Wikimedia org admins at

I think people should think carefully before getting involved in this or other "education" programs, because they stray from the Wikipedia ideal of universal knowledge. Specifically, the Google Code-In is only available to people from 13 to 17 who are enrolled in schools and submit their personal details to Google and sign the following agreement. We should be inspired to keep a better to-do list of coding tasks, and help people to do them, but those people should not be restricted to that portal. (Yes, I recognize that this education initiative is actually not as bad as other course-oriented programs that are only available to paying students at a particular college, but we should keep both at arm's length) Wnt (talk) 07:21, 19 November 2013 (UTC)

By the way, if a 13-year-old can handle a task like this, including running the JavaScript unit tests, he's pretty fucking awesome. His next special assignment ought to be to come back to us and tell us how to do it. Wnt (talk) 07:37, 19 November 2013 (UTC)

I will not try to convince you about the usefulness of these programs. :) Let me say though that we are not limiting participation to anybody. We do or best supporting any volunteers willing to work on technical tasks. We find these programs useful to reach out to new contributors, but we ask their participants to use our community channels just like anybody else: Bugzilla, Gerrit, etc. Some of those contributors stay around and become regular community members.

On a more specific note, is there a list of wikitext modules waiting to be rewritten in Lua? Even if it's only one template, I would be willing to create a Google Code-in task for it and see what happens. Thank you for your help!--

talk

) 20:56, 19 November 2013 (UTC)

PS: That task you link to has been filed by Bartosz Dziewoński aka

talk

) 21:00, 19 November 2013 (UTC)

I've started a project To-do list at Wikipedia:Lua/To do. There are only three templates listed there now, but there are a lot of templates that need converting. I'll try and expand the list some more later, but other editors should of course feel free to add to it as well. — Mr. Stradivarius ^{♪ talk ♪} 23:47, 19 November 2013 (UTC)

Thank you

talk

) 16:31, 21 November 2013 (UTC)

Yes Qgil, I'd be happy to mentor Lua-related stuff. Let me know what I need to do to get mentoring set up, and let's get started. :) — Mr. Stradivarius ^{♪ talk ♪} 22:02, 21 November 2013 (UTC)

talk

) 19:26, 22 November 2013 (UTC)

Well, if the kids want to come here I and others will be happy to see if we can help, as for anyone else, no registration required. Wnt (talk) 03:18, 23 November 2013 (UTC)

Sorry for the late reply, GCI is being intense. It would be good to have at least one mentor to bridge with the community here. I can take the risk of being the mentor myself as long as I can rely on you for quick revisions of the work submitted (within 24 hours, 36 max). What we are doing is create bug reports for GCIU tasks, CCing there community contributors with related skills, and we encourage the students to ask and report progress there, allowing anybody to help regardless of GCI involvement. We will requests students to create templates and modules under the Sandbox here. There you can watch the changes and help testing them. You really don't need any GCI registration to participate in any if these activities, but still I would appreciate one registered mentor in order to avoid the risk of me becoming the bottleneck. I still volunteer to create the tasks, which is probably the most boring part.  :) Thank you for your understanding and your support. I hope it pays off, as it is paying off to other areas that are getting involved in GCI.--

talk

) 02:19, 28 November 2013 (UTC)

I have created two Google Code-in tasks to rewrite Template:Interval and Template:Countdown, based on the useful information provided at

talk

) 00:50, 7 December 2013 (UTC)

talk

) 06:03, 8 December 2013 (UTC)

There's a nice discussion about rewriting {{Multiple image}} template to Lua. It needs a good height solution and border parameter options. --Rezonansowy (talk • contribs) 19:06, 7 December 2013 (UTC)

As was mentioned in the discussion, there is no mechanism for Lua to determine the image's height or aspect ratio at present, so a rewrite won't be able to help. isaacl (talk) 00:27, 8 December 2013 (UTC)

Really? Is there any Lua programmer, who can confirm this or solve? --Rezonansowy (talk • contribs) 17:27, 8 December 2013 (UTC)

Isaacl is correct that there isn't a way to get access to that data right now, but as pointed out in the discussion you linked, there has been work going on at bugzilla:41498 that might change that in the not-so-distant future. There was also another way around the issue that was pointed out, but I would prefer to see the outcome of the Bugzilla discussion before writing a module based on the workaround. We don't want to have to go to the effort of writing a module, only to have to rewrite it from scratch after we get a new feature. — Mr. Stradivarius ^{♪ talk ♪} 06:38, 10 December 2013 (UTC)

I don't understand. According to Wikipedia:Extended image syntax you can set an image height directly using "xheightpx". Why are they talking about calculating according to the aspect ratio? Just set them all to be 100 px high or whatever. Wnt (talk) 08:16, 10 December 2013 (UTC)

My understanding of the discussion is that the width of the enclosing box has to be set, but I haven't looked at the implementation. isaacl (talk) 09:10, 10 December 2013 (UTC)

I have just finished writing Module:Arguments, a module that uses metatables to fetch arguments from the frame and parent frame objects. I made it because I was getting tired of copying and pasting the same argument-processing code into all of my modules, and also because I wanted a way to process arguments that didn't use the pairs function by default. It can accept values from the frame and the parent frame at the same time, and there are quite a few configuration options. Test cases are at Module:Arguments/testcases (results). Have a look and see what you think, and let me know if you find any issues or want any new features. And of course, feel free to try it out on your own modules. :) — Mr. Stradivarius ^{♪ talk ♪} 06:48, 10 December 2013 (UTC)

Thanks! I was going to suggest that after your work on Module:Redirect. Clearly it's good to have the code in one place and maximally updated.

The key thing I don't understand is: why do you go straight for "if frame == mw.getCurrentFrame()"? You don't even use the getCurrentFrame for anything, and it's expensive to call. How do you end up having the module called with the "wrong" frame so that you don't want to use the arguments it provides?

If there's a reason why you need to check the frame is the right one, I think you should handle the logic for type(frame) == 'table' [or nil?] first, and only go for the getCurrentFrame when you know you need to look at it. Wnt (talk) 08:05, 10 December 2013 (UTC)

The reason I used mw.getCurrentFrame was to tell whether the module is being called from #invoke, or from the debug console or another Lua module. If we only check whether frame is a table, it would cause errors if frame is an argument table passed straight from Lua, as we might not have frame.args or frame.getParent. I suppose you're right, though - if we check that we have frame, frame.args, and frame.getParent, we can be reasonably sure that we are dealing with a frame object without using getCurrentFrame. On the other hand, if the function is being called from Lua, it will pretty much always be overkill to use Module:Arguments, so it is best to write modules that have a function specially for Lua (as recommended at Wikipedia:Lua style guide#Naming conventions). I'll put an alternative version in the sandbox, and wait to see what others have to say on the subject. — Mr. Stradivarius ^{♪ talk ♪} 08:52, 10 December 2013 (UTC)

I've long disliked your use of frame == mw.getCurrentFrame(), as it makes it impossible to use mw.getCurrentFrame():newChild{} to pass in an actual frame object for testing from the debug console. Anomie⚔ 16:46, 10 December 2013 (UTC)

Anomie, you should have said something! I'll never learn to fix my bad code if you don't tell me what's wrong with it. :) For my part, I simply copied the practice from the modules that were already present on Wikipedia - I think it might have been Toohool who came up with it first. But anyway, we now have a decision - mw.getCurrentFrame is out, and type checking is in. If you have any other implementation tips or things that you think could be improved, I'd love to hear them - this module will probably end up being used in a lot of other modules, after all. (I'm already thinking of using it in Module:Math, and Theopolisme has already used it in Module:Multiple issues.) — Mr. Stradivarius ^{♪ talk ♪} 21:48, 10 December 2013 (UTC)

It never bothered me enough to track down who did it. But then you outed yourself ;) Anomie⚔ 22:18, 10 December 2013 (UTC)

Code review:

• I'd raise an error if options isn't a table, instead of silently ignoring it. Same for frame. BTW, mw:Extension:Scribunto/Lua_reference_manual#libraryUtil.
• __newindex should also throw an error if it doesn't succeed, not silently fail.
• I wonder if you really need mw.ustring.find( val, '%S' ) instead of string.find; in general, you're only going to see ASCII whitespace in blank arguments and string.find is going to be faster. For that matter, string.gsub( val, '^%s*(.-)%s*$', '%1' ) should be faster than mw.text.trim.
• I'd like to see an explicit return value from tidyVal in all code paths.
• Memoizing nils from tidyVal could be useful. Stick local nilArg = {} at the top of the script and use that to indicate nil in metaArgs. Or just use false if you don't care about supporting false in luaArgs.
• Instead of using argTables, I might just have options.frameOnly assign {} to pargs, options.parentOnly assign {} to fargs, and have options.parentFirst swap them (fargs, pargs = pargs, fargs). Then hard-code the search order in __index and mergeArgs. Although maybe that's going a bit overboard, the extra time for iterating over argTables is probably negligible.
• On the topic of over-optimization, I might well choose one of four versions of tidyVal based on options, instead of checking the options on each call.

HTH. Anomie⚔ 23:36, 10 December 2013 (UTC)

Thanks for the tips! I've implemented all of them apart from the one about eliminating argTables. I did try it, but the only way to do it without introducing extra functions or if statements is to repeat the exact same code three times. And that applies to both the both __index and mergeArgs functions. It just looks too ugly to me, and reminds me of coding templates again. :) Also, I think LuaJIT might do some loop unrolling anyway? — Mr. Stradivarius ^{♪ talk ♪} 13:18, 11 December 2013 (UTC)

I don't know if LuaJIT unrolls loops, but I do know that Scribunto doesn't use LuaJIT. It turns out that stock Lua is plenty fast for our needs. Anomie⚔ 13:53, 11 December 2013 (UTC)

Ah, for some reason I thought it did. In that case, it looks like we have a choice between either fast code or sane-looking code. I have a slight preference for sane-looking code, but I'd like to hear what others think as well. — Mr. Stradivarius ^{♪ talk ♪} 14:38, 11 December 2013 (UTC)

Unless there is a significant real-world performance problem, easier-to-maintain code is the way to go, for the sake of everyone who may have to understand the code in future (often including the code's author in six months :-). isaacl (talk) 16:25, 11 December 2013 (UTC)

Quoting myself: "Although maybe that's going a bit overboard, the extra time for iterating over argTables is probably negligible." Go with the clear version here, IIRC I tested it and the difference was around 0.3 microseconds. Anomie⚔ 20:21, 11 December 2013 (UTC)

I'm working on Module:Math/sandbox at the moment, and I have a question about the greatest common divisor function. If you know about that kind of thing, I'd be grateful if you could comment over at Module talk:Math#Testcases. Thanks. :) — Mr. Stradivarius ^{♪ talk ♪} 03:43, 12 December 2013 (UTC)

So I've been working on a new feature at Module:Random to generate a random date. While doing so I've noticed that math.random seems to be limited to 32-bit integers. For example, these are the first few outputs for math.random(9999999999999999):

• 1874484855
• 1260195387
• 626812935
• 470599561
• 1048108424
• 456570891
• 628875670
• 1019583173
• 520927378
• 1472467616
• 656070341
• 485208199
• 1807948255
• 1610626572
• 283570239

I would expect higher numbers than that, and the cutoff point seems to be 2^32 / 2. And from poking around the Internet, this does seem to be a known issue.[8][9] Does anyone know any ways that this can be worked around, and should we file a bug about it? — Mr. Stradivarius ^{♪ talk ♪} 14:10, 25 November 2013 (UTC)

A quick and dirty way is to just generate and concatenate individual random digits until hitting the given digit count. Then, wrap the number (take away the maximum/ceiling until it is below it) if it is larger than allowed. —  HELLKNOWZ  ▎TALK 14:58, 25 November 2013 (UTC)

Hm, that will work well for positive integers, but I'm going to need negative integers as well. I essentially need to call math.random(-someReallyBigNumber, otherReallyBigNumber). The long negative numbers occur when somebody wants to make a random date before about 1950 or so and it gets converted to a Unix timestamp. — Mr. Stradivarius ^{♪ talk ♪} 15:08, 25 November 2013 (UTC)

A very dirty way would be to calculate the relative difference of the floor/ceiling numbers (say, -300 to -100 is 3:1). Then choose which number (negative or positive) to generate based on that (75% vs 25%). (Only problem is if the ratio exceeds int32, so the random to choose which one would get botched up.)

Another "simpler" way is to shift the floor to 0, generate the random number as before, then shift it back. —  HELLKNOWZ  ▎TALK 15:14, 25 November 2013 (UTC)

Also, if you multiply the random's result by how much larger the ceiling is than the random's int32 limitation, then add a random number up to random ceiling, it might work and give proper distribution (assuming you don't go above lua max int values, which is int64?). Say, for simplicity, max random allows is 1234, but you need up to 123456. It's 123456/1234~=100 time bigger. Multiply random roll by 100, then add 0..100 to it. So 536 * 100 + 78 = 53678. (This might have corner cases on the very highest roll...) —  HELLKNOWZ  ▎TALK 15:21, 25 November 2013 (UTC)

The cutoff point is RAND_MAX, which on modern systems is typically 2^31-1. If you need numbers in a greater range, you could try something like this:

function myRandom( l, u )
    local r = math.random() + math.random() / 2147483648
    return math.floor( r * ( u - l + 1 ) ) + l
end

Or if you don't want to assume RAND_MAX is 2^31-1, you could sum math.random( 0, N ) / ( N ^ x ) for x from 1 to ceiling( 53 / N ). If you're wanting a range (u - l + 1) larger than 2^53 with unit precision, you'll need more complicated math. The exponent of 53 comes from the fact that Lua uses 64-bit floating point internally, so it can represent all integers up to 2^53 but 2^53+1 == 2^53. HTH. Anomie⚔ 18:00, 25 November 2013 (UTC)

Thank you Anomie, that worked magnificently. :) (You probably guessed that already from your barnstar though.) I have to admit, I didn't fully understand your example of "sum math.random( 0, N ) / ( N ^ x ) for x from 1 to ceiling( 53 / N )". What does N refer to in this instance? — Mr. Stradivarius ^{♪ talk ♪} 03:12, 28 November 2013 (UTC)

It would help if I hadn't screwed up the formula. Should be "ceiling( 53 / log₂(N) )". N is some number, ideally a power of 2 (so log₂(N) is an integer). Anomie⚔ 03:20, 28 November 2013 (UTC)

I'm not sure I understand why math.random()/2147483648 is being added to r; if I understand correctly, since math.random() is between 0 and 1, dividing it by 2147483648 is going to result in a negligibly small amount, and r is essentially just math.random().

Mr. Stradivarius, note that with this approach, not every number within your specified range (if it spans a range greater than RAND_MAX) can be produced. As long as this is not an issue, then this method is fine. isaacl (talk) 05:41, 28 November 2013 (UTC)

It would be nice to have all the numbers in the specified range available, but it's not of vital importance in the current implementation. It might become more important in the future depending on what people want to do with the module. I need to read up on the maths we're using here, because it's years since I studied this kind of thing and I fear that I have forgotten most of it. If anyone knows a good starting place, that would be helpful. :) Also, I coded up Anomie's algorithm in Module:User:Mr. Stradivarius/sandbox, but it seems to be returning numbers greater than 1 for small values of N. Is that just because I need to use big numbers, or is it another problem? Yours mathematically, — Mr. Stradivarius ^{♪ talk ♪} 07:48, 28 November 2013 (UTC)

Not that I'm pushing my method, but if you only need up to int64 with all numbers included, then here's a simple(r) random as I mentioned above: Module:Sandbox/Hellknowz/Misc, giving stuff for max=21474836480000 like 1176639272617. —  HELLKNOWZ  ▎TALK 09:59, 28 November 2013 (UTC)

@Isaacl: math.random() returns numbers between 0 and 1, but the number will always be a rational number i/2147483647 for some integer i. Lua can handle rational numbers in the range 0–1 with denominators up to 9007199254740992 (2^53). So math.random()/2147483648 isn't negligibly small. Although now that I look at it more carefully, that should be math.random()/2147483647 to be absolutely correct. Anomie⚔ 14:15, 28 November 2013 (UTC)

@Mr. Stradivarius: Bah, I made another error. Use math.random( 0, N - 1 ), not math.random( 0, N ), since the range is inclusive when you specify arguments to math.random. Anomie⚔ 14:15, 28 November 2013 (UTC)

@Hellknowz: If you try out your method, I think you'll find that it never returns an odd number greater than 2^53, never a non-multiple of 4 over 2^54, and so on. Lua just can't represent those numbers exactly. Also, it blows up if you pass it a max greater than 4611686016279904512 (that's 2^62 - 2^31 + 2^8). Anomie⚔ 14:15, 28 November 2013 (UTC)

To be honest that's more thought than I ever gave it this morning. I'm sure it breaks horribly somewhere before or at 2147483648*2147483648 because the ratio exceeds random's limit. I did mention this is only for up to around int64. It's not really meant to output anything so large, I'm just upping the lua's random by a power of 2 or whatever the limit and time variable size is before it goes all exponent-y. —  HELLKNOWZ  ▎TALK 14:43, 28 November 2013 (UTC)

However, it seems to me that an r between 0 and 1 is all that is needed, not an r that is the sum of one random value from 0 to 1 and another from 0 to 1/2147483648. By extending the distribution, the result can go beyond the desired upper limit (though just by a small fraction of the range; I said it was negligible not because Lua couldn't handle the precision, but because the frequency of seeing values above the desired limit would be relatively small). OK, I understand now: the first random number will never be 1, nor have a precision finer than 1/2147483648, so the second random number is filling in precision below 1/2147483648.

There is a bug, though, where the function could rarely give 1 due to rounding errors. So perhaps the best simple replacement function is math.random( 0, 67108863 ) / 67108864 + math.random( 0, 134217727 ) / 9007199254740992. Anomie⚔ 17:37, 28 November 2013 (UTC)

On another note for Mr. Stradivarius: not sure if a uniform distribution is required; adding two uniform distributions together results in a triangular distribution. isaacl (talk) 16:36, 28 November 2013 (UTC)

I haven't quite worked through Anomie's second solution, so I'm not sure if the following proposal is the same: you can partition the desired range into sections < RAND_MAX, randomly choose a section, then randomly choose a number within that section. This will result in all values being possible with a uniform distribution. However, assuming each section is of size N, there will usually be one section left over at the end with a smaller size, and so the probability of choosing this section has to be weighted accordingly. If I get the chance, I'll try to create a demo. isaacl (talk) 16:43, 28 November 2013 (UTC)

Your proposal appears to be equivalent. The general form of my (corrected) functions is ${\displaystyle {\frac {r_{1}}{b_{1}}}+{\frac {r_{2}}{b_{1}b_{2}}}}$, with ${\displaystyle 0\leq r_{1}<b_{1}}$ and ${\displaystyle 0\leq r_{2}<b_{2}}$. Since the maximum value of the second term, ${\displaystyle {\frac {b_{2}-1}{b_{1}b_{2}}}={\frac {1}{b_{1}}}-{\frac {1}{b_{1}b_{2}}}}$, is less than the separation between values in the first term (${\displaystyle {\frac {1}{b_{1}}}}$), the first term is serving as your "partition the desired range into sections" and the second is serving as "choose a number within that section". Of course, this assumes that the "desired range" is expressable as [0,b₁b₂) for two integers b₁ and b₂. By taking the largest usable range (2⁵³ = 2²⁶2²⁷), dividing it down to [0,1), and multiplying it up to the desired range, the error in the distribution is probably going to be comparable to the error in the distribution of math.random() itself (which, here, is implemented as [10] and (probably) [11]). Anomie⚔ 17:37, 28 November 2013 (UTC)

OK, here's my version: Module:Sandbox/isaacl/Probability (testcases) The algorithm is as follows:

• Let random() return a value in the range [0, 1), and random(N) return a value in the range [0, N).
• Given a range [0, M): divide it into T partitions, with T-1 partitions of size N, and the remaining partition covering the rest of the range. Thus T = ceiling(M/N).
  • Partition 0 range: [0, N)
  • Partition 1 range: [N, 2N)
  • ...
  • Partition (T-1) range: [(T-1)*N, M)
• Select a partition s = floor(random() * M / N)
  • If s is in the range 0 to T-2: result = s*N + floor(random(N))
  • If s is T-1: result = s*N + floor(random(M-(T-1)*N))

This approach results in a uniform distribution (assuming random() is uniform). isaacl (talk) 03:23, 29 November 2013 (UTC)

I am not an expert of randomness. But I know that devising a good random generator is not easy. IMO, the reference source is Knuth's book The Art of Computer Programming, chap. 3, vol. 2, Random numbers. I believe that you could find there the answer to every question that you are looking for. D.Lazard (talk) 00:35, 18 December 2013 (UTC)

Many infobox templates have a |native_name_lang= parameter, which is supposed to take an ISO 639-1 value such as "cs". As can be seen where I fixed this example, people sometimes enter instead text values, such as "Czech".

Could we have a module, please, to either i) throw an error message in such cases, and add a tracking category or ii) substitute the ISO value in the rendered HTML? If we go for the former, I'll then request a BOT to fix up existing instances. Andy Mabbett (Pigsonthewing); Talk to Andy; Andy's edits 22:52, 6 December 2013 (UTC)

@Pigsonthewing: I don't think substituting the ISO value is possible without creating a database of all known language codes and their various English names. (While finding a name from a language code is now easy thanks to the latest update to Scribunto's language library, I don't believe it is yet possible to find the language code from the name.) However, it is trivial to detect what is a valid language code and what isn't, so I chose to go with your option number one. Take a look at Template:Check ISO 639-1 and see if it does what you want. At the moment it is set to only accept language codes that are recognised by the MediaWiki software, but if that turns out to be too strict then we can relax that criterion without too much trouble. Let me know if it needs tweaking at all. — Mr. Stradivarius ^{♪ talk ♪} 06:32, 13 December 2013 (UTC)

On the point of recognising language codes, it does recognise all of the codes listed in the ISO 639-1 language matrix - see the test case results. — Mr. Stradivarius ^{♪ talk ♪} 09:53, 13 December 2013 (UTC)

I've removed that particular part from the test cases, as now the module just uses a local table, so you can see what codes are supported from the module page. — Mr. Stradivarius ^{♪ talk ♪} 15:47, 13 December 2013 (UTC)

Note that Scribunto has been able to get the English name (or the name in many other languages) for a particular MediaWiki language code for a long time now using mw.language.fetchLanguageName. The recent addition allows for fetching the entire list of codes (and corresponding names). Anomie⚔ 14:04, 13 December 2013 (UTC)

I'm obviously having a bad day... I'll need to investigate this further. Possibly the best thing would be to have a hybrid solution, where language names that are recognised are automatically converted to an ISO 639-1 language code. But that will have to wait for tomorrow. — Mr. Stradivarius ^{♪ talk ♪} 15:45, 13 December 2013 (UTC)

I see that mw.language.fetchLanguageNames isn't deployed to the English Wikipedia yet, so we might have to wait a little while for a hybrid solution. It's working on test2wiki though, so I can start work on the coding at least - tomorrow. :) — Mr. Stradivarius ^{♪ talk ♪} 15:58, 13 December 2013 (UTC)

Thank you for your efforts. Should I now hold off until you've done more? Andy Mabbett (Pigsonthewing); Talk to Andy; Andy's edits 18:33, 13 December 2013 (UTC)

That depends how much red error text you can put up with in articles. If you wait for the hybrid solution we should be able to convert most English language names to ISO 639-1 codes, but if you deploy now all of those will come up as errors. Also, I would appreciated someone double-checking the list of ISO 639-1 codes at Module:Check ISO 639-1 to see if I have missed any or made any mistakes. — Mr. Stradivarius ^{♪ talk ♪} 01:31, 14 December 2013 (UTC)

The code is now working over at test2wiki:Module:Check ISO 639-1. All we need to do now is wait for mw.language.fetchLanguageNames to be made available on enwiki, and it should be ready to deploy. — Mr. Stradivarius ^{♪ talk ♪} 07:05, 14 December 2013 (UTC)

Thank you again. There's no great rush; and I'm loathe to introduce red error messages unnecessarily. Do we have an ETA for mw.language.fetchLanguageNames? Is it weeks, months or years? Also, do we need (or have you already included) a tracking category? Andy Mabbett (Pigsonthewing); Talk to Andy; Andy's edits 20:28, 14 December 2013 (UTC)

Not sure about the schedule, but I suspect days or weeks rather than months, and definitely not years. And yes, I have included a tracking category: Category:Pages with invalid ISO 639-1 language codes. — Mr. Stradivarius ^{♪ talk ♪} 00:24, 15 December 2013 (UTC)

The schedule is at mw:MediaWiki 1.23/Roadmap. Since the change was merged on Dec 10, it was included in 1.23wmf7 (which is why it's already on testwiki). Anomie⚔ 00:30, 15 December 2013 (UTC)

Ok, so it should be on enwiki on Thursday 19th. Thanks Anomie. :) — Mr. Stradivarius ^{♪ talk ♪} 01:04, 15 December 2013 (UTC)

@Pigsonthewing: {{Check ISO 639-1}} is now converting language names to ISO 639-1 codes, so feel free to roll it out to infoboxes. Merry Christmas. :) — Mr. Stradivarius ^{♪ talk ♪} 06:24, 25 December 2013 (UTC)

And Merry Christmas to you, too. That looks great; thank you. It's now deployed in {{Infobox settlement}}. I wonder whether a template performing the reverse, (en -> English, de-> German) would also be useful? Andy Mabbett (Pigsonthewing); Talk to Andy; Andy's edits 21:53, 26 December 2013 (UTC)

Hi everyone. Today I created Module:TableTools, which includes a few table-related functions that I noticed that I was writing over and over again. It includes functions for getting the union and intersection of n tables, functions for getting numbers of table keys, and functions for dealing with sparse arrays. Test cases are at Module:TableTools/testcases (run tests). Let me know if you can see any improvements that can be made or any new features you would like to be added. — Mr. Stradivarius ^{♪ talk ♪} 15:52, 15 December 2013 (UTC)

Some comments:

• I wonder whether the general loop-over-args is better as you have it or as something like this:

function foo( ... )
    for i = 1, select( '#', ... ) do
        local t = select( i, ... )
        -- Do stuff with t
    end
end

• Why name the functions like "getUnion" rather than just "union"?
• Why do union and intersection sort?
  • Also note that the sort is going to blow up if ret contains mixed types or incomparable types.
• getAffixNums should construct the pattern string outside the loop and store it in a local, instead of reconcatenating each time.
• Note your union and intersection functions will blow up if one of the arrays contains NaN ;)

HTH Anomie⚔ 18:33, 15 December 2013 (UTC)

Thank you as always for your very useful comments. :) I have put all of your suggestions to use. It is probably easiest to respond in order:

• I had forgotten about the select function, as I haven't actually used it in a module before. After reading the docs I was debating whether to handle nil values for union and intersect, but in the end I decided that it wasn't important.
• I have developed a habit of making all my function names verbs, but I agree that this isn't really necessary here.
• I actually put union and intersect in as a kind of afterthought. I added the sparse array functions first as I used those the most, and then thought that I should add union and intersection as they often appear in sets of table functions, and I had seen union used in Module:Infobox. So the table sort was also kind of an afterthought, but I can now see that it's not a good idea. People can always sort the tables later if they need to.
• Good catch. I had copied most of that function from a version I wrote before, and forgot to check for inefficiencies inside the loop.
• I've hacked union and intersect to not blow up on NaN. It uses a NaN proxy variable similar to the proxy variable for nils in Module:Arguments that you suggested above. Converting back to NaN is ugly, though - all the NaNs become 0/0. Basically I am assuming that it doesn't matter how a NaN is generated, only that it is a NaN, but I'm still a relative NaN noob so I might have this wrong.

I'm thinking about also adding union and intersection functions for table key/value pairs. If anyone has a favourite template function that they would like to be included, then now is the time to ask - it will be easier to do it now than when the module is heavily transcluded. — Mr. Stradivarius ^{♪ talk ♪} 14:33, 16 December 2013 (UTC)

Not sure if these are the type of utilities you had in mind, but a clone function and a display function would be useful (see Module:Sandbox/isaacl/ColourSpace/Tuple for sample implementations for the case where tables are being used as a 1-dimensional vector). isaacl (talk) 15:35, 16 December 2013 (UTC)

We already have mw.clone and mw.logObject - are these the kind of thing that you were looking for? — Mr. Stradivarius ^{♪ talk ♪} 21:45, 16 December 2013 (UTC)

For cloning, mw.clone should be fine; for display, I was looking for something more tailored for displaying vectors, with a configurable separator (as the function in my module does). isaacl (talk) 00:19, 17 December 2013 (UTC)

Do you need something that will work over tables recursively? If not, table.concat and mw.text.listToText should do the job just fine. If we need something that works recursively with configurable separators then we can always try adapting the mw.logObject source code, but I'm not sure how much such a function would be used. — Mr. Stradivarius ^{♪ talk ♪} 01:50, 17 December 2013 (UTC)

Module:Sandbox/isaacl/ColourSpace was designed to allow new colour format specifications to be defined in separate modules. The colour data is stored within a vector, so to avoid having each format re-implement a display function using table.concat, I made a common utility in Module:Sandbox/isaacl/ColourSpace/Tuple. Math-oriented applications could benefit from having a common display utility, but I agree that I don't know how much use this would get. isaacl (talk) 02:08, 17 December 2013 (UTC)

• Don't know it is in your scope for this, but here are two.

• Could the module provide a 2D textual table, and by extension a 2D table in wiki/HTML table format? There is table.concat that produces table[i] .. sep .. table[i+1] .. sep .. table[j] as a stream (string), and one separator in all.

Usefulness? A bare 2D textual (like with /n : after each pair, or a sepEndOfPair option) I would surely use in development time, checking a status. A wikitable may be too simple for mainspace too (e.g., one wants headers & colheaders from day 1).

(Fixed layout: this was meant to be idea #1 for MR Strad). -DePiep (talk) 14:30, 24 December 2013 (UTC)

Sorry, missed this one. In TableTools, I'd like to avoid anything that generates wikitext, as I see it as being limited to pure Lua table operations. Outputting a Lua table as a wikitable sounds like it might be useful, but it is probably best left for a different module. Maybe Module:Wikitable? As for outputting a Lua table to text, I have got by fine up to now with table.concat, mw.text.listToText, mw.logObject, and Module:User:Anomie/deepToString. I can't really think of any generalised application that isn't covered by those, but if there is one then maybe we could add it. (Maybe in a separate module?) — Mr. Stradivarius ^{♪ talk ♪} 14:41, 24 December 2013 (UTC)

Thanx. Clear. -DePiep (talk) 16:58, 24 December 2013 (UTC)

• How does the TableTool work with deeper tables? (3D, table-in-table values, ...)? I have plans to become a novice in this (yes), so don't spend your time explaining to me before I do some homework ;-). Just to drop the idea. -DePiep (talk) 09:52, 20 December 2013 (UTC)

At the moment, none of the functions in TableTools work with nested tables. It's intended to be for fairly simple, commonly used operations on tables that would be a pain to rewrite every time. Operations on nested tables often need to be customised to the application in question, so it would be hard to write something general enough. As for displaying nested tables for development purposes, that already exists: see mw.logObject. We also have Module:User:Anomie/deepToString if you ever want something that displays outside of the debug console. — Mr. Stradivarius ^{♪ talk ♪} 14:21, 24 December 2013 (UTC)

• Adding a third suggestion: provide a procedure from wikitable to Lua data page. This could circumvent needing an off-wiki (off-line) script to transform a wikitable into a Lua /data page. (Currently done for module:convert/data). It could work like this:

1. I have made a sound wikitable in my userspace. 2. I copy-paste that table code into a module subpage (that will look bad, posing as Lua code). 3. Proposed TableTools function wikitableToDatapage reads that datapage as input (a single long string?). It parses the text, and returns a correct Lua data page (Lua data table). 4. One invokes that function in say userspace, and the result page showing can be copied into the true module:x/data page.

It could work, and imo serves a need. But it would make very few people proud & happy I guess.-DePiep (talk) 07:32, 24 December 2013 (UTC)

That could be useful, yes. I think it would work best as a JavaScript tool rather than a Lua module, however. If it was in JavaScript you wouldn't have to worry about having to save the page to convert the text - you could do it right in the edit window. Also, you would be able to use it on pages other than module pages (maybe on documentation pages). With Lua it would only be possible to input the data directly on a module page, due having to escape pipes and equals signs etc. in MediaWiki template syntax. — Mr. Stradivarius ^{♪ talk ♪} 14:21, 24 December 2013 (UTC)

OK. -DePiep (talk) 14:30, 24 December 2013 (UTC)

• I'd made some blundering efforts at Module:DisplayLuaTableContents, and again just now; the less said of these the better, but my general intent is: we should have some robust functions keys(table) -> list of all table keys, values(table) -> list of all table values, flat(table) -> some limited-depth digest of everything in the table that can be used to generate a tostring(table) -> showing "everything" in it. These are complicated by the possibility of having looped tables and complicated situations like a = {1, 2}; a[1] = a, etc. But I think that practically, as a debugging tool, from absolute beginner to whatever level of knowledge I have now, the age-old question of whatinhell is really IN that table seems to rank always at the top of the list. Wnt (talk) 22:44, 20 December 2013 (UTC)

  What you're describing here sounds an awful lot like mw.logObject - mw.logObject can't display just keys or just values, but it does a very good job of displaying all the content in a table. Have you tested mw.logObject out, and does it do what you want? — Mr. Stradivarius ^{♪ talk ♪} 14:21, 24 December 2013 (UTC)

A good suggestion. I haven't explored the logging features as much as I ought. Wnt (talk) 18:08, 30 December 2013 (UTC)

I've knocked together Module:FindFeatures (and Module:FindFeatures/displayDatabase to produce the data files for it to use) to list adjacent features to those on various celestial bodies. A sample usage is at Addams (crater), which links to the sandbox version of a template (which links to the sandbox version of another) that calls it. Anyway, it wasn't inspired programming, and it would affect at least hundreds of articles when active, which I'm starting to think about doing, so I thought I should see if anyone is interested in looking it over. Thanks! Wnt (talk) 17:51, 30 December 2013 (UTC)

Talk:List of Hindu festivals#2014 dates

Please see the above. -- 签名 _sig at 09:51, 31 December 2013 (UTC)

Category:Fixed holidays (on non-Gregorian calendars) includes most holidays and observances based on calendars other than the Gregorian. These could benefit from some sort of a calendar function. Some basic calendar stuff is currently done in Template:Infobox holiday/date. -- 签名 _sig at 10:58, 1 January 2014 (UTC)

I guess I am missing a very simple point here. About {{convert}}. {{hands/sandbox}} invokes module:convert/sandbox. When there are 3 unnamed parameter inputs, it produces a range output (from-to-to, ok). When there is only one input parameter (the number), unnamed parameter #2 should simply pass on "hands" (a length unit) in the invoke. But I can't get it working. Anyone?

See testcases. Edit {{hands/sandbox}} if you like. -DePiep (talk) 06:08, 29 December 2013 (UTC)

@DePiep: I think the problem is that Module:Convert takes configuration parameters from the invoking template, and convert parameters from the values passed to the invoking template. Currently, {{hands/sandbox}} is using "#invoke" to directly call the sandbox module, and that means the numbered parameters that it defines are ignored. The fix would be to call the template {{Convert/sandbox}} rather than the module. Johnuniq (talk) 00:57, 30 December 2013 (UTC)

Tanks, too busy IRL to explore this. -DePiep (talk) 11:18, 3 January 2014 (UTC)

Still don't get it. I don´t see why the sandboxes should behave fundamentally different from the live pages (module and template).

{{convert}} has no named params set by itself, and #invokes module:convert.

{{convert/sandbox}} today has two named params set by itself, and #invokes module:convert-sandbox. It passed on parameters as expected (named and unnnamed params in the calling template). DePiep (talk) 14:38, 9 January 2014 (UTC)

Is this issue a module:convert only thing, or standard module behaviour?

Module:Convert is using this capability in a way that most other modules don't, but it is a feature available to all modules, yes. I've written an explanation of this issue at Module:Arguments#Frames and parent frames which might prove useful. It's mostly geared towards explaining how frames and parent frames are used in that particular module, but the general concept extends to other modules as well. Let me know if it helps. — Mr. Stradivarius ^{♪ talk ♪} 15:29, 9 January 2014 (UTC)

Will have to study this. Mind that it should answer my question: how or why do #invokes of module:sandbox and module:convert/sandbox behave differently in this parameter handling (I can not simply replace them for invoking). -DePiep (talk) 15:42, 9 January 2014 (UTC)

@DePiep: Consider a hypothetical Module:M and Template:T. Suppose Template:T has {#invoke:M|main|warnings=2} and someone puts {T|text=hello} in an article.

The main function in Module:M will be run, and that function has access to two completely different sets of arguments. The article provides the "parent" frame which has arguments "text=hello", while the template provides a frame with arguments "warnings=2".

The module can do whatever it wants with each set of arguments. Convert uses the template arguments as configuration options that apply site-wide, while the article arguments provide the details for the particular conversion.

Other modules may have no need for configuration options, and may merge arguments from both sources, then operate on the result. For example, if the module adds numbers, using {#invoke:M|main|11|22|33} might result in 66, and using {T|11|22|33} might do exactly the same. For that to occur, the module has to merge the frame's arguments and the parent frame's arguments (and decide what to do if different arguments are provided by each). Johnuniq (talk) 21:25, 9 January 2014 (UTC)

Can anyone work out what is causing the script error at Template:Orphan/sandbox? I get the following traceback:

Lua error in Module:HtmlBuilder at line 41: attempt to concatenate field 'val' (a nil value).

Backtrace:

    Module:HtmlBuilder:41: ?
    (tail call): ?
    Module:HtmlBuilder:54: ?
    (tail call): ?
    Module:HtmlBuilder:25: ?
    [C]: in function "tostring"
    Module:Message_box/sandbox:539: ?
    (tail call): ?
    (tail call): ?
    mw.lua:558: ?
    [C]: in function "getExpandedArgument"
    mw.lua:264: ?
    Module:Unsubst:13: in function "chunk"
    mw.lua:558: ?

I've worked out what causes that particular error in HtmlBuilder: it's when you pass a table to .cssText, as I have done in my module sandbox here. I've also worked out that args.style on line 207 of Module:Message box/sandbox is an empty table, when it should be nil. This makes me suspect that it is a nasty bug in Module:Arguments, as I used empty tables to memoize nil values in it. However, I can't seem to reproduce the issue. Can anyone see what I'm missing? — Mr. Stradivarius ^{♪ talk ♪} 19:48, 12 January 2014 (UTC)

The bug is in Module:Arguments's __pairs and __ipairs: you can't just iterate over metaArgs, because it contains these table-for-nil entries. This should fix it. Anomie⚔ 21:00, 12 January 2014 (UTC)

D'oh, I had a feeling it was something like that. Thanks for the fix. :) — Mr. Stradivarius ^{♪ talk ♪} 05:18, 13 January 2014 (UTC)

I was just reading WP:Editnotice and realized that on every single page we are running Template:Editnotice load and perhaps many of its bastard progeny.

I understand though that converting template to module doesn't guarantee speed improvement because there's an overhead, and there are many other technical considerations regarding parser functions/magic words/etc. Should we discuss, though, whether it is possible to shave some time off all our page loads this way? — Preceding unsigned comment added by Wnt (talk • contribs) 17:06, 19 January 2014 (UTC)

Note that it's loaded on the edit pages themselves, not parses or page views. Anomie⚔ 17:30, 19 January 2014 (UTC)

My understanding from WP:Editnotice is that editnotices can be loaded on page view, as on MediaWiki:Common.css, but are turned off by an explicit test "{{#ifeq: {{{notice action|}}} | view | ... | ...}}". Which means that the template itself is loaded on all page views, doesn't it? Wnt (talk) 17:51, 19 January 2014 (UTC)

Interesting, I didn't know people did that. But no, it's not all page views, just ones that include MediaWiki:Clearyourcache (and there there's a test to only show it in the MediaWiki namespace), which is only CSS and JS pages. Anomie⚔ 18:43, 19 January 2014 (UTC)

Thanks! That makes it less disturbing, though I suppose Wikid77 would care even if it's only for edits :) I'm afraid I said something to the contrary under this misunderstanding at WT:Drafts lately, where they're looking for a pagenotice for the entire Draft namespace. If you have any ideas to help them out (and correct my misinformation) I'm sure they'd be grateful. ;) Wnt (talk) 19:14, 19 January 2014 (UTC)

I don't think there is any decent way to auto-add a banner to pages. Your best bet would probably be to get whoever set up the hack for allowing anon page creation in the draft namespace to add a hack that uses the BeforePageDisplay hook to prepend some message to all pages in that namespace. Anomie⚔ 20:21, 19 January 2014 (UTC)

Template:Demography seems like a good candidate to Lua-ise; it currently relies on 1000+ #if statements to check for every year 1500–2100, which I think could be shortened and simplified easily with Lua. ~HueSatLum 22:54, 17 January 2014 (UTC)

Now that's how we wrote template code in the good old days. ;) This template is such a good example of how needlessly complicated template code can get that I took a screenshot for posterity. I imagine we could make that many times smaller and many times faster with Lua. — Mr. Stradivarius ^{♪ talk ♪} 04:17, 18 January 2014 (UTC)

I do not understand the purpose of the template as writing simple table is much easier than any template or Lua code for that task. Wouldn't be better to use bot that convert all places where the template is used to simple table? Lua is very useful if the number of arguments may vary, and there are some additional aggregated results provided from the template. See for example pl:Module:Demografia I wrote some time ago on Polish Wikipedia. Paweł Ziemian (talk) 15:28, 18 January 2014 (UTC)

You might be right about that. But nevertheless, I've created an initial version at Module:Demography. It still needs to be made compatible with HTML 5, and it needs some documentation and test cases, but the basics are there. — Mr. Stradivarius ^{♪ talk ♪} 13:43, 19 January 2014 (UTC)

Also, at the moment it is 3,401 bytes compared to Template:Demography's 44,350. And when I compared the speeds in my sandbox with 100 template invocations, the total page load time with Template:Demography was 34.903 seconds, while with Module:Demography it was 1.157 seconds. — Mr. Stradivarius ^{♪ talk ♪} 13:55, 19 January 2014 (UTC)

Wow, much improved! Thank you! ~HueSatLum 15:54, 19 January 2014 (UTC)

It's now HTML-5-ified and being used live in articles. — Mr. Stradivarius ^{♪ talk ♪} 03:36, 20 January 2014 (UTC)

It looks good, but I'm disappointed - where's the HTML5? I see tr, td, th ... I'm still waiting on that inline SVG. Also, looking at the result for Tirana it seems like some lines are missing ... there's no really obvious reason in the code popping out at me, and it's variable - I move it up and down and the bottom line appears and disappears while I scroll! I get separate cells for 1955 1985 and 2001 if I delete the 1703 cell. It's not somehow scaled so that they're less than a full pixel - looks like a 1px border - on inspection the cells seem slightly overlapping, but one of them should get a border... it's mysterious. Wnt (talk) 05:51, 20 January 2014 (UTC) doh, sorry, I had the browser zoomed out by a step! And apparently it doesn't handle it very predictably. Wnt (talk) 19:05, 20 January 2014 (UTC)

By "HTML-5-ified", I mean "doesn't fail HTML 5 validation". If you want to go crazy with inline SVGs, be my guest. :) — Mr. Stradivarius ^{♪ talk ♪} 21:42, 20 January 2014 (UTC)

Well, the problem is: 1 <canvas> 2 <audio> 3 <video> ... 5 <embed> 6 <track> 7 <datalist> 8 <keygen> 9 <output> 10 <article> 11 <aside> 12 <bdi> 13 <command> 14 <details> 15 <dialog> 16 <summary> 17 <figure> 18 <figcaption> 19 <footer> 20 <header> 22 <meter> 23 <nav> 24 <progress> .. 29 <time> 30 <wbr> - that's 23/24 of the first 24 HTML5 elements that aren't recognized (list from [12]). Another is <syntaxhighlight lang=""> which is a wiki tag interpreted in a way that has nothing to do with its widely recognized use in HTML5. However, the last part of the list is looking a little more promising. The mark tag seems to work ... I haven't looked into how 25 26 27 works and I'm not sure if they are just being taken out or if they're being interpreted. Ditto 10:10 in "time" tags. But xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxwbr:xxxxxxxxxxxxxcontrol<control>xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx breaks for me at the wbr, so that must be enabled. And definitely the 28 <section> tag can foul up a preview, it appears. Still, the vast majority are not available, though ... I haven't tried every which way to try to hack unauthorized HTML tags through Lua yet. 😈 Wnt (talk) 23:40, 20 January 2014 (UTC)

It seems like most of the

My first thought is that this would be a task better suited for a bot than for a module. With a module you would have to go through all four million redirects and add a new template to them, or appropriate the {{

WP:BOTR and seeing what people say. — Mr. Stradivarius ^{♪ talk ♪}

05:43, 22 January 2014 (UTC)

I could only find mw.title thingie and a Lua thingie to retrieve number of pages in a category. Is there a way to, either with existing Lua functionality or by expanding it, to retrieve fresh category members using Lua? I'm thinking of doing it by fetching data from api URLs even. (In my understanding, if such Lua module is invoked from a template, its content is not refreshed until a contributor clicks 'purge'; so, while expensive, such function would be called rarely.) --Gryllida 04:35, 22 January 2014 (UTC)

No, it's not currently possible to retrieve any information about individual category members using Lua. This would have to be implemented as a new feature in Scribunto. Also, I don't think it's possible to directly get information about recent category members with the API, only total category members, although there are plenty of bots that monitor categories over time and deduce which pages are new. (Better get someone who knows the API better than me to confirm that.) As for updating categories that are transcluded from a template, purge won't do it - you need a

null edit to the categorised page or the equivalent API call. — Mr. Stradivarius ^{♪ talk ♪}

05:23, 22 January 2014 (UTC)

maybe not as such, but you can get "timestamp" property with the "categorymembers" API, where the timestamp mark when this page was added to the cat, so you can get "all", and filter the "fresh" yourself. would be more efficient to ask for "fresh" directly (btw: why?), but since you can get up to 500 members in a call, in most cases you'll get it with a single call, or small number of "continue"s. see [13]. peace - קיפודנחש (aka kipod) (talk) 06:22, 22 January 2014 (UTC)

actually, looking at it more carefully, you can probably ask for "fresh" also, through "cmstart" parameter...[14]. peace - קיפודנחש (aka kipod) (talk) 06:24, 22 January 2014 (UTC)

With Lua it seems that it is possible with mw:Extension:DynamicPageListEngine but it is not installed on Wikipedias. --Vriullop (talk) 10:49, 22 January 2014 (UTC)

קיפודנחש (aka kipod)The API can do this, but can you help get URL contents using Lua? Socket in Lua module doesn't appear to work, with the usual generic error message. --Gryllida 23:34, 23 January 2014 (UTC)

For reference, I'm including current version of the module and the invoke call here.

Module:

-- loads the HTTP module and any libraries it requires

local http = require("socket.http")

 

local p = {}

 

function p.main(frame)

  r, e = http.request("http://wrong.host/")

  return e

end

 

return p

Invoke:

{{#invoke:Foo|main}}

--Gryllida 23:54, 23 January 2014 (UTC)

Gryllida: ttbomk, the lua "socket.http" module is not available on wikimedia wikis. if you maintain your own wiki (with scribunto installed), i would not recommend installing/enabling it (and i wouldn't be able to advise how to do such a thing), unless you have very strict control over who can create modules. peace - קיפודנחש (aka kipod) (talk) 00:48, 24 January 2014 (UTC)

קיפודנחש (aka kipod) Thanks. A safer workaround you could suggest? Gryllida 01:00, 24 January 2014 (UTC)

Access to MW API discussion, which links to a mailing list discussion; frame:preprocess (used in Module:Subpages) may be of interest, as well as Bug 60386, which I just filed. I expect it to be a formal step to record interest in such feature, with pending feedback from the relevant people. --Gryllida 00:59, 24 January 2014 (UTC)

Thanks all. This route appears to be unsafe in its current shape, with Lua modules being writable and invokable by quite all contributors, but gadgets — the existing way to talk to API visible to all contributors — being writable only by sysops (and user scripts being limited in scope to one contributor only). --Gryllida 02:02, 24 January 2014 (UTC)

Break 1: User rights

User:BJorsch_(WMF) קיפודנחש (aka kipod)With Scribunto, can a separate namespace (or other measure) be applied to some Lua modules, so they...

• ... are editable by sysops (or other user rights group members) only, and...
  • (Similar to how JS (it can query API or URLs) on page load for all users is though gadgets, which only sysops edit. --Gryllida 01:59, 24 January 2014 (UTC))
• ...have nice things such as access to sockets and/or API?

--Gryllida 01:54, 24 January 2014 (UTC)

I think that we should be careful about moving in this direction. There are a whole lot of ways that trusted developers can introduce new features in any project. I think the noble goal of Scribunto is to put powerful scripting tools into the hands of users, so that a wide range of creative ideas are obtained, and a larger number of people pick up coding skills. The introduction of powerful libraries into Scribunto has helped make that happen, and I wouldn't want to see that good work slacken as people decide it's easier just to make the functions available with less caution to a smaller group. Wnt (talk) 06:08, 24 January 2014 (UTC)

This is not going to fly. I'm hoping that mw.api will be developed to make more data available for modules, but it has to be done in a manner that is compatible with the purpose of the servers that render pages—that purpose is to provide encyclopedic content without undue delay or expense. Tools to do lengthy calculations with categories or to access data from a network need to be run on dedicated servers, such as

WP:LABS. Johnuniq (talk

) 06:33, 24 January 2014 (UTC)