User:Iritscen/PageCountAudit

This page looks into the math that is used by MediaWiki to give various page counts.

Magic words

These first two magic words provide easy answers to the amount of content on the wiki, but are they accurate and useful?

NUMBEROFPAGES: 4,642

MW code says: Simply count all entries in page table.

Iritscen says: Though problematic in the past, this magic word now matches the grand total (see below) of all PAGESINNS counts, including files and redirects. However, we don't really want to display that catch-all number on our main page.

NUMBEROFARTICLES: 901

MW code says: If the count method global is set to 'link', the software gets a distinct count of the entries in the pagelinks table, "pl_from" field, that match those page ids. In other words, it filters out pages that do not link to other pages (the reasoning presumably being that "those aren't real wiki pages" if they're not connecting to anything else). It also filters out redirects. If the method is set to 'comma', it counts all non-blank pages (yes, really).

Iritscen says: Okay, $wgArticleCountMethod has now been set to 'comma'. (Note: As of MW 1.31, "The 'comma' value for $wgArticleCountMethod is no longer supported for performance reasons, and installations with this setting will now work as if it was configured with 'any'." It appears that 'any' will return the same result, counting all pages that are not redirects.)

PAGESINNS, AKA PAGESINNAMESPACE: These counts agree with the number of pages displayed for each namespace on the Special:AllPages page, which provides some much-needed verifiability. However, since Allpages counts redirect pages, that means that PAGESINNS does too. Therefore, we can't use a straight sum of PAGESINNS results as our page count. See final section for the adjusted number.

PAGESINNS breakdown:

Namespace	ID	Page count
Media	-2	<not available>
Special	-1	<not available>
Main	0	749
Talk	1	113
User	2	123
User talk	3	51
OniGalore	4	15
OniGalore talk	5	2
File	6	2,665
File talk	7	17
MediaWiki	8	39
MediaWiki talk	9	1
Template	10	131
Template talk	11	6
Help	12	2
Help talk	13	2
Category	14	191
Category talk	15	6
BSL	100	60
BSL talk	101	6
OBD	102	191
OBD talk	103	36
AE	104	23
AE talk	105	15
Oni2	108	37
Oni2 talk	109	18
XML	110	125
XML talk	111	18

All articlespaces (without File) totalled using PAGESINNS: 1686

All talkspaces totalled using PAGESINNS: 274

All contentspaces (as currently defined in $wgContentNamespaces = {0, 2, 100, 102, 104, 108, 110}) totalled using PAGESINNS: 1308

The grand total for all namespaces (including File) is: 4642

Redirects

There were 353 redirects as of 2021-12-13 according to Special:ListRedirects.

Redirect breakdown:

Main	280
Talk	0
User	0
User talk	0
OniGalore	2
OniGalore talk	0
File	0
File talk	0
MediaWiki	0
MediaWiki talk	0
Template	0
Template talk	0
Help	1
Help talk	1
Category	0
Category talk	0
BSL	2
BSL talk	0
OBD	18
OBD talk	0
AE	0
AE talk	0
Oni2	2
Oni2 talk	0
XML	46
XML talk	0

Conclusion

NUMBEROFPAGES is too broad to be useful, but now that the page-count method is 'comma' (or 'all'), I am able to reconcile NUMBEROFARTICLES with PAGESINNS. PAGESINNS in turn reconciles with AllPages, which lists each page onscreen and is thus verifiable by a direct count (which I have done in the past). So to see how the math works out, we can get the directly-verifiable count by using PAGESINNS on all "content" namespaces, and then manually subtracting redirects as counted above.

Namespaces Main, User, BSL, OBD, XML, AE, and Oni2 totaled using PAGESINNS: 1308

Same total minus redirects in those namespaces: 960

At the time of this writing (2021-12-13), this gives a total "content minus redirect" total of 919, which is 29 higher than the value (890) returned by NUMBEROFARTICLES. That's an acceptable margin of error, though one that could bear some investigation (I would be curious if the page-count method omits blank or very short pages, which we have a number of). Thus NUMBEROFARTICLES is fit for use on the main page.