Page MenuHomePhabricator
Create Task
Maniphest T388985

Add support for more complex linking in <math/> formulas
Open, Needs TriagePublicFeature
Actions

Description

Currently, the <math /> tag in MediaWiki supports linking to other articles only for very simple attributes. For example, in the formula a + b = c^d, links can be created for individual elements like a, b, c, and d, but not for more complex expressions like c^d or a + b.

Related: T134311: Allow for links in <math/> tags

Details

Related Changes in Gerrit:
SubjectRepoBranchLines +/-
Add support for more complex linking in <math/> formulasmediawiki/extensions/Mathmaster+249 -45
Fix performance testsmediawiki/extensions/MathSearchmaster+6 -1
Customize query in gerrit

Related Objects
Search...

Event Timeline

FrederikHennecke1 created this task.Mar 15 2025, 9:32 PM
Restricted Application added a subscriber: Aklapper. · View Herald TranscriptMar 15 2025, 9:32 PM
gerritbot added a comment.Mar 15 2025, 9:39 PM

Change #1128019 had a related patch set uploaded (by FrederikHennecke1; author: FrederikHennecke1):

[mediawiki/extensions/Math@master] Add support for more complex linking in <math/> formulas

https://gerrit.wikimedia.org/r/1128019

gerritbot added a project: Patch-For-Review.Mar 15 2025, 9:39 PM
Aklapper changed the subtype of this task from "Task" to "Feature Request".Mar 16 2025, 8:55 AM
Physikerwelt added a project: RemexHtml.Mar 20 2025, 8:13 AM
Physikerwelt subscribed.

I think we should be using RemexHtml for math and implement tree matching.

FrederikHennecke1 added a comment.Sep 13 2025, 1:20 AM

I propose the following changes for this task:

Commit 1: Move linking logic to MMLbase
Currently, we have the core logic for linking in MathNativeMML. We should move it to MMLbase so that we can use the caching in the second commit.

Changes:

  • Refactor link-handling code into MMLbase class
  • Preserve existing linking functionality without optimizations
  • Add basic tests for MMLbase linking methods

Commit 2: Implement LocalChecker caching with links
Right now, only the base MathML is getting cached, and whenever we need to add the linking, we have to recompute it.

Changes:

  • Modify LocalChecker to inject links into MathML after cache lookup
  • Add tests verifying:
    • Link persistence through caching
    • Correct link rendering in checker output
    • Cache hit/miss behavior with links

Commit 3: Optimize with subtree hashing + Firefox fix
Currently, the <math /> tag in MediaWiki supports linking only for very simple attributes. For example, in the formula a + b = c^d, links can be created for individual elements like a`, b, c, and d but not for more complex expressions like c^d or a + b. I propose we use subtree hashing to find the correct subtree and add a new MMLanchor (just a normal <a href=...>) element to it.

Changes:

  • Implement subtree hashing in MMLbase
  • Use hashes for faster link/formula matching (Merkle tree hashing)
  • Add hash collision tests in MMLbase tests
  • While we are at it, add a fix for T402358
Physikerwelt added a comment.Sep 13 2025, 3:04 PM

Thank you that’s a great plan. However I think it’s most effective if we discuss the caching and rendering question a bit more in detail before proceeding with the implementation.

  1. do we have current measurements of the performance and the impact of caching?
  1. can we estimate how the proposed changes would affect the performance? (Now, we have almost no qid / vs future with x % qid)
  1. can we generate a code pen, with a few complex and simple examples that demonstrates how the rendering will look like when implementation is done. That would help us to test the impact on even individual browsers.
Physikerwelt added a comment.Sep 14 2025, 8:37 AM

One of the open questions for me is how to do linking. There are quite complex, MathML core-compliant solutions that use a tags. However, this requires the addition of a foreign tag to the MathML tree (as you demonstrated in your previous commit). Instead, I'd recommend being a bit more flexible and just use href from the full spec and use a polyfill (JavaScript) for the linking without introducing a new element to the DOM. I think this would be much more elegant, and the downside that people without JS and with browsers that only support core won't get links. I think that's acceptable. At least the href element is supported by FF, Safari https://developer.mozilla.org/en-US/docs/Web/MathML/Reference/Global_attributes/href (and MathJax)

FrederikHennecke1 added a comment.Sep 14 2025, 4:47 PM
In T388985#11178481, @Physikerwelt wrote:

Thank you that’s a great plan. However I think it’s most effective if we discuss the caching and rendering question a bit more in detail before proceeding with the implementation.

  1. do we have current measurements of the performance and the impact of caching?
  1. can we estimate how the proposed changes would affect the performance? (Now, we have almost no qid / vs future with x % qid)
  1. can we generate a code pen, with a few complex and simple examples that demonstrates how the rendering will look like when implementation is done. That would help us to test the impact on even individual browsers.
  1. I don't think so. We only store strings instead of the MMLbase tree as that's the recommended way: https://www.mediawiki.org/wiki/Object_cache . If we have to add the QID, then we have to either recompute the MMLbase tree for subtree matching to add the anchor tag or do subtree matching in the DOM, which is more complicated.
  1. Can't really say as I don't know much about the caching performance itself.
  1. It should be easy if it is only about rendering. We could add anchor tags in random placed in the formula tree and then compare it with the default rendering.
In T388985#11178870, @Physikerwelt wrote:

One of the open questions for me is how to do linking. There are quite complex, MathML core-compliant solutions that use a tags. However, this requires the addition of a foreign tag to the MathML tree (as you demonstrated in your previous commit). Instead, I'd recommend being a bit more flexible and just use href from the full spec and use a polyfill (JavaScript) for the linking without introducing a new element to the DOM. I think this would be much more elegant, and the downside that people without JS and with browsers that only support core won't get links. I think that's acceptable. At least the href element is supported by FF, Safari https://developer.mozilla.org/en-US/docs/Web/MathML/Reference/Global_attributes/href (and MathJax)

This sounds like a good idea. How about: instead of adding a new anchor tag, like in https://gerrit.wikimedia.org/r/c/mediawiki/extensions/Math/+/1180252, we just add the href (and maybe title) to the correct MathML elements and add a polyfill to make the links work in chromium and other browsers.

gerritbot added a comment.Sep 16 2025, 1:24 PM

Change #1188775 had a related patch set uploaded (by Physikerwelt; author: Physikerwelt):

[mediawiki/extensions/MathSearch@master] Fix performance tests

https://gerrit.wikimedia.org/r/1188775

Physikerwelt added a comment.Sep 16 2025, 1:26 PM

This sounds like a good idea. How about: instead of adding a new anchor tag, like in https://gerrit.wikimedia.org/r/c/mediawiki/extensions/Math/+/1180252, we just add the href (and maybe title) to the correct MathML elements and add a polyfill to make the links work in Chromium and other browsers.

This seems the best solution to me.

gerritbot added a comment.Sep 18 2025, 8:33 AM

Change #1188775 had a related patch set uploaded (by Physikerwelt; author: Physikerwelt):

[mediawiki/extensions/MathSearch@master] Fix performance tests

https://gerrit.wikimedia.org/r/1188775

gerritbot added a comment.Sep 18 2025, 9:00 AM

Change #1188775 merged by jenkins-bot:

[mediawiki/extensions/MathSearch@master] Fix performance tests

https://gerrit.wikimedia.org/r/1188775

Physikerwelt added a comment.Sep 18 2025, 12:01 PM

Re-caching. I'll investigate. My feeling is that we might get around caching altogether. For the average case, the most time-consuming phase is the checking phase. Here, we might be able to improve performance just by updating the library. We could not do that before, because the new version requires PHP 8. However, a typical rendering time for a random formula is about 1ms. Some exceptional cases such as

\begin{pmatrix}



360	&	0,032	&	-0,012	&	0,980	&	0,00001	&	0,00633	&	0,00005	&	-0,00002	&	0,00155	&	0,17559	&	0,00535	&	0,81907	\\
365	&	0,031	&	-0,012	&	0,981	&	0,00001	&	0,00617	&	0,00005	&	-0,00002	&	0,00159	&	0,17526	&	0,00520	&	0,81953	\\
370	&	0,030	&	-0,012	&	0,982	&	0,00001	&	0,00600	&	0,00005	&	-0,00002	&	0,00164	&	0,17494	&	0,00506	&	0,82000	\\
375	&	0,029	&	-0,012	&	0,983	&	0,00002	&	0,00583	&	0,00011	&	-0,00004	&	0,00337	&	0,17565	&	0,00603	&	0,81831	\\
380	&	0,027	&	-0,012	&	0,984	&	0,00004	&	0,00549	&	0,00019	&	-0,00009	&	0,00699	&	0,17398	&	0,00463	&	0,82139	\\
385	&	0,027	&	-0,012	&	0,984	&	0,00006	&	0,00549	&	0,00031	&	-0,00014	&	0,01147	&	0,17398	&	0,00463	&	0,82139	\\
390	&	0,026	&	-0,011	&	0,985	&	0,00012	&	0,00614	&	0,00051	&	-0,00022	&	0,01927	&	0,17375	&	0,00517	&	0,82108	\\
395	&	0,026	&	-0,011	&	0,986	&	0,00022	&	0,00615	&	0,00092	&	-0,00039	&	0,03481	&	0,17374	&	0,00517	&	0,82109	\\
400	&	0,025	&	-0,011	&	0,986	&	0,00040	&	0,00597	&	0,00166	&	-0,00073	&	0,06541	&	0,17343	&	0,00502	&	0,82155	\\
405	&	0,024	&	-0,011	&	0,987	&	0,00064	&	0,00580	&	0,00265	&	-0,00121	&	0,10886	&	0,17311	&	0,00488	&	0,82201	\\
410	&	0,022	&	-0,011	&	0,988	&	0,00121	&	0,00546	&	0,00488	&	-0,00244	&	0,21898	&	0,17247	&	0,00459	&	0,82294	\\
415	&	0,021	&	-0,010	&	0,990	&	0,00218	&	0,00612	&	0,00749	&	-0,00356	&	0,35287	&	0,17224	&	0,00513	&	0,82263	\\
420	&	0,018	&	-0,010	&	0,991	&	0,00400	&	0,00560	&	0,01287	&	-0,00715	&	0,70839	&	0,17129	&	0,00470	&	0,82401	\\
425	&	0,014	&	-0,008	&	0,993	&	0,00730	&	0,00653	&	0,01564	&	-0,00894	&	1,10942	&	0,17021	&	0,00548	&	0,82431	\\
430	&	0,009	&	-0,005	&	0,996	&	0,01160	&	0,00812	&	0,01286	&	-0,00714	&	1,42317	&	0,16892	&	0,00679	&	0,82429	\\
435	&	0,001	&	-0,001	&	0,999	&	0,01684	&	0,00998	&	0,00169	&	-0,00169	&	1,68502	&	0,16679	&	0,00833	&	0,82488	\\
440	&	-0,008	&	0,005	&	1,004	&	0,02300	&	0,01332	&	-0,01382	&	0,00863	&	1,73380	&	0,16457	&	0,01105	&	0,82439	\\
445	&	-0,021	&	0,012	&	1,009	&	0,02980	&	0,01676	&	-0,03734	&	0,02134	&	1,79425	&	0,16121	&	0,01384	&	0,82495	\\
450	&	-0,039	&	0,022	&	1,017	&	0,03800	&	0,02178	&	-0,06804	&	0,03838	&	1,77424	&	0,15666	&	0,01785	&	0,82549	\\
455	&	-0,062	&	0,034	&	1,027	&	0,04800	&	0,02757	&	-0,10796	&	0,05920	&	1,78826	&	0,15084	&	0,02241	&	0,82675	\\
460	&	-0,091	&	0,052	&	1,039	&	0,06000	&	0,03719	&	-0,14683	&	0,08390	&	1,67648	&	0,14397	&	0,02985	&	0,82618	\\
465	&	-0,128	&	0,076	&	1,052	&	0,07390	&	0,05027	&	-0,18816	&	0,11172	&	1,54641	&	0,13550	&	0,03978	&	0,82472	\\
470	&	-0,182	&	0,117	&	1,064	&	0,09098	&	0,07415	&	-0,22330	&	0,14355	&	1,30545	&	0,12408	&	0,05755	&	0,81837	\\
475	&	-0,258	&	0,184	&	1,074	&	0,11260	&	0,11524	&	-0,25209	&	0,17978	&	1,04940	&	0,10969	&	0,08692	&	0,80339	\\
480	&	-0,366	&	0,290	&	1,076	&	0,13900	&	0,18226	&	-0,27912	&	0,22116	&	0,82060	&	0,09138	&	0,13244	&	0,77617	\\
485	&	-0,519	&	0,457	&	1,063	&	0,16930	&	0,29072	&	-0,30224	&	0,26613	&	0,61904	&	0,06905	&	0,20083	&	0,73012	\\
490	&	-0,714	&	0,699	&	1,015	&	0,20800	&	0,45230	&	-0,32835	&	0,32145	&	0,46677	&	0,04552	&	0,29486	&	0,65962	\\
495	&	-0,945	&	1,024	&	0,920	&	0,25860	&	0,67444	&	-0,36234	&	0,39263	&	0,35275	&	0,02350	&	0,41279	&	0,56372	\\
500	&	-1,166	&	1,389	&	0,777	&	0,32300	&	0,93034	&	-0,40482	&	0,48224	&	0,26976	&	0,00848	&	0,53836	&	0,45317	\\
505	&	-1,349	&	1,744	&	0,606	&	0,40730	&	1,18453	&	-0,46385	&	0,59967	&	0,20837	&	0,00046	&	0,65707	&	0,34247	\\
510	&	-1,335	&	1,930	&	0,405	&	0,50300	&	1,33598	&	-0,50263	&	0,72665	&	0,15248	&	0,01412	&	0,74997	&	0,23591	\\
515	&	-1,205	&	1,968	&	0,237	&	0,60820	&	1,38807	&	-0,52798	&	0,86230	&	0,10384	&	0,03921	&	0,81202	&	0,14877	\\
520	&	-0,981	&	1,852	&	0,129	&	0,71000	&	1,33233	&	-0,52278	&	0,98693	&	0,06874	&	0,07462	&	0,83386	&	0,09152	\\
525	&	-0,737	&	1,665	&	0,072	&	0,79320	&	1,22298	&	-0,47800	&	1,07988	&	0,04670	&	0,11445	&	0,82615	&	0,05940	\\
530	&	-0,515	&	1,475	&	0,040	&	0,86200	&	1,10758	&	-0,40081	&	1,14796	&	0,03113	&	0,15486	&	0,80561	&	0,03953	\\
535	&	-0,330	&	1,310	&	0,020	&	0,91490	&	1,00606	&	-0,30010	&	1,19130	&	0,01819	&	0,19295	&	0,78149	&	0,02556	\\
540	&	-0,170	&	1,162	&	0,008	&	0,95400	&	0,91401	&	-0,17744	&	1,21284	&	0,00835	&	0,22979	&	0,75409	&	0,01612	\\
545	&	-0,029	&	1,028	&	0,001	&	0,98030	&	0,83003	&	-0,03425	&	1,21412	&	0,00118	&	0,26586	&	0,72430	&	0,00983	\\
550	&	0,098	&	0,905	&	-0,003	&	0,99500	&	0,75253	&	0,12958	&	1,19659	&	-0,00397	&	0,30184	&	0,69257	&	0,00560	\\
555	&	0,212	&	0,792	&	-0,004	&	1,00000	&	0,68090	&	0,31135	&	1,16317	&	-0,00587	&	0,33731	&	0,65885	&	0,00383	\\
560	&	0,318	&	0,688	&	-0,005	&	0,99500	&	0,61515	&	0,51436	&	1,11283	&	-0,00809	&	0,37364	&	0,62441	&	0,00196	\\
565	&	0,411	&	0,593	&	-0,004	&	0,97860	&	0,55445	&	0,72542	&	1,04665	&	-0,00706	&	0,40859	&	0,58931	&	0,00209	\\
570	&	0,497	&	0,507	&	-0,004	&	0,95200	&	0,49980	&	0,94667	&	0,96572	&	-0,00762	&	0,44393	&	0,55483	&	0,00123	\\
575	&	0,575	&	0,428	&	-0,003	&	0,91540	&	0,44943	&	1,17115	&	0,87175	&	-0,00611	&	0,47865	&	0,51983	&	0,00152	\\
580	&	0,645	&	0,358	&	-0,002	&	0,87000	&	0,40496	&	1,38568	&	0,76911	&	-0,00430	&	0,51204	&	0,48604	&	0,00192	\\
585	&	0,707	&	0,295	&	-0,002	&	0,81630	&	0,36475	&	1,58223	&	0,66019	&	-0,00448	&	0,54467	&	0,45412	&	0,00121	\\
590	&	0,762	&	0,240	&	-0,002	&	0,75700	&	0,32981	&	1,74901	&	0,55087	&	-0,00459	&	0,57533	&	0,42413	&	0,00054	\\
595	&	0,809	&	0,193	&	-0,001	&	0,69490	&	0,29995	&	1,87422	&	0,44712	&	-0,00232	&	0,60249	&	0,39627	&	0,00124	\\
600	&	0,847	&	0,154	&	-0,001	&	0,63100	&	0,27499	&	1,94353	&	0,35337	&	-0,00229	&	0,62669	&	0,37256	&	0,00075	\\
605	&	0,880	&	0,121	&	-0,001	&	0,56680	&	0,25402	&	1,96354	&	0,26999	&	-0,00223	&	0,64823	&	0,35147	&	0,00030	\\
610	&	0,906	&	0,095	&	-0,001	&	0,50300	&	0,23750	&	1,91879	&	0,20120	&	-0,00212	&	0,66585	&	0,33420	&	-0,00006\\
615	&	0,926	&	0,074	&	-0,001	&	0,44120	&	0,22398	&	1,82404	&	0,14577	&	-0,00197	&	0,68048	&	0,31988	&	-0,00036\\
620	&	0,942	&	0,058	&	0,000	&	0,38100	&	0,21382	&	1,67849	&	0,10335	&	0,00000	&	0,69104	&	0,30812	&	0,00084	\\
625	&	0,955	&	0,045	&	0,000	&	0,32100	&	0,20556	&	1,49128	&	0,07027	&	0,00000	&	0,70052	&	0,29882	&	0,00065	\\
630	&	0,965	&	0,035	&	0,000	&	0,26500	&	0,19921	&	1,28370	&	0,04656	&	0,00000	&	0,70793	&	0,29156	&	0,00051	\\
635	&	0,973	&	0,027	&	0,000	&	0,21700	&	0,19413	&	1,08765	&	0,03018	&	0,00000	&	0,71393	&	0,28567	&	0,00040	\\
640	&	0,980	&	0,021	&	0,000	&	0,17500	&	0,19049	&	0,90031	&	0,01929	&	0,00000	&	0,71849	&	0,28120	&	0,00031	\\
645	&	0,985	&	0,015	&	0,000	&	0,13820	&	0,18650	&	0,72990	&	0,01112	&	0,00000	&	0,72305	&	0,27673	&	0,00022	\\
650	&	0,989	&	0,011	&	0,000	&	0,10700	&	0,18396	&	0,57525	&	0,00640	&	0,00000	&	0,72612	&	0,27371	&	0,00016	\\
655	&	0,992	&	0,008	&	0,000	&	0,08160	&	0,18205	&	0,44463	&	0,00359	&	0,00000	&	0,72844	&	0,27144	&	0,00012	\\
660	&	0,994	&	0,006	&	0,000	&	0,06100	&	0,18078	&	0,33540	&	0,00202	&	0,00000	&	0,72999	&	0,26992	&	0,00009	\\
665	&	0,995	&	0,005	&	0,000	&	0,04458	&	0,18015	&	0,24623	&	0,00124	&	0,00000	&	0,73077	&	0,26916	&	0,00007	\\
670	&	0,997	&	0,004	&	0,000	&	0,03200	&	0,17969	&	0,17755	&	0,00071	&	0,00000	&	0,73155	&	0,26839	&	0,00006	\\
675	&	0,997	&	0,003	&	0,000	&	0,02320	&	0,17888	&	0,12931	&	0,00039	&	0,00000	&	0,73232	&	0,26763	&	0,00004	\\
680	&	0,998	&	0,002	&	0,000	&	0,01700	&	0,17824	&	0,09519	&	0,00019	&	0,00000	&	0,73310	&	0,26687	&	0,00003	\\
685	&	0,999	&	0,001	&	0,000	&	0,01192	&	0,17761	&	0,06705	&	0,00007	&	0,00000	&	0,73388	&	0,26610	&	0,00001	\\
690	&	1,000	&	0,000	&	0,000	&	0,00821	&	0,17697	&	0,04639	&	0,00000	&	0,00000	&	0,73467	&	0,26533	&	0,00000	\\
695	&	1,000	&	0,000	&	0,000	&	0,00572	&	0,17697	&	0,03234	&	0,00000	&	0,00000	&	0,73467	&	0,26533	&	0,00000	\\
700	&	1,000	&	0,000	&	0,000	&	0,00410	&	0,17697	&	0,02318	&	0,00000	&	0,00000	&	0,73467	&	0,26533	&	0,00000	\\
705	&	1,000	&	0,000	&	0,000	&	0,00293	&	0,17697	&	0,01655	&	0,00000	&	0,00000	&	0,73467	&	0,26533	&	0,00000	\\
710	&	1,000	&	0,000	&	0,000	&	0,00209	&	0,17697	&	0,01182	&	0,00000	&	0,00000	&	0,73467	&	0,26533	&	0,00000	\\
715	&	1,000	&	0,000	&	0,000	&	0,00148	&	0,17697	&	0,00839	&	0,00000	&	0,00000	&	0,73467	&	0,26533	&	0,00000	\\
720	&	1,000	&	0,000	&	0,000	&	0,00105	&	0,17697	&	0,00592	&	0,00000	&	0,00000	&	0,73467	&	0,26533	&	0,00000	\\
725	&	1,000	&	0,000	&	0,000	&	0,00074	&	0,17697	&	0,00418	&	0,00000	&	0,00000	&	0,73467	&	0,26533	&	0,00000	\\
730	&	1,000	&	0,000	&	0,000	&	0,00052	&	0,17697	&	0,00294	&	0,00000	&	0,00000	&	0,73467	&	0,26533	&	0,00000	\\
735	&	1,000	&	0,000	&	0,000	&	0,00036	&	0,17697	&	0,00204	&	0,00000	&	0,00000	&	0,73467	&	0,26533	&	0,00000	\\
740	&	1,000	&	0,000	&	0,000	&	0,00025	&	0,17697	&	0,00141	&	0,00000	&	0,00000	&	0,73467	&	0,26533	&	0,00000	\\
745	&	1,000	&	0,000	&	0,000	&	0,00017	&	0,17697	&	0,00097	&	0,00000	&	0,00000	&	0,73467	&	0,26533	&	0,00000	\\
750	&	1,000	&	0,000	&	0,000	&	0,00012	&	0,17697	&	0,00068	&	0,00000	&	0,00000	&	0,73467	&	0,26533	&	0,00000	\\
755	&	1,000	&	0,000	&	0,000	&	0,00009	&	0,17697	&	0,00048	&	0,00000	&	0,00000	&	0,73467	&	0,26533	&	0,00000	\\
760	&	1,000	&	0,000	&	0,000	&	0,00006	&	0,17697	&	0,00034	&	0,00000	&	0,00000	&	0,73467	&	0,26533	&	0,00000	\\
765	&	1,000	&	0,000	&	0,000	&	0,00004	&	0,17697	&	0,00024	&	0,00000	&	0,00000	&	0,73467	&	0,26533	&	0,00000	\\
770	&	1,000	&	0,000	&	0,000	&	0,00003	&	0,17697	&	0,00017	&	0,00000	&	0,00000	&	0,73467	&	0,26533	&	0,00000	\\
775	&	1,000	&	0,000	&	0,000	&	0,00002	&	0,17697	&	0,00012	&	0,00000	&	0,00000	&	0,73467	&	0,26533	&	0,00000	\\
780	&	1,000	&	0,000	&	0,000	&	0,00002	&	0,17697	&	0,00008	&	0,00000	&	0,00000	&	0,73467	&	0,26533	&	0,00000	\\
785	&	1,000	&	0,000	&	0,000	&	0,00001	&	0,17697	&	0,00006	&	0,00000	&	0,00000	&	0,73467	&	0,26533	&	0,00000	\\
790	&	1,000	&	0,000	&	0,000	&	0,00001	&	0,17697	&	0,00006	&	0,00000	&	0,00000	&	0,73467	&	0,26533	&	0,00000	\\
795	&	1,000	&	0,000	&	0,000	&	0,00001	&	0,17697	&	0,00006	&	0,00000	&	0,00000	&	0,73467	&	0,26533	&	0,00000	\\
800	&	1,000	&	0,000	&	0,000	&	0,00001	&	0,17697	&	0,00006	&	0,00000	&	0,00000	&	0,73467	&	0,26533	&	0,00000	\\
805	&	1,000	&	0,000	&	0,000	&	0,00001	&	0,17697	&	0,00006	&	0,00000	&	0,00000	&	0,73467	&	0,26533	&	0,00000	\\
810	&	1,000	&	0,000	&	0,000	&	0,00001	&	0,17697	&	0,00006	&	0,00000	&	0,00000	&	0,73467	&	0,26533	&	0,00000	\\
815	&	1,000	&	0,000	&	0,000	&	0,00001	&	0,17697	&	0,00006	&	0,00000	&	0,00000	&	0,73467	&	0,26533	&	0,00000	\\
820	&	1,000	&	0,000	&	0,000	&	0,00001	&	0,17697	&	0,00006	&	0,00000	&	0,00000	&	0,73467	&	0,26533	&	0,00000	\\
825	&	1,000	&	0,000	&	0,000	&	0,00001	&	0,17697	&	0,00006	&	0,00000	&	0,00000	&	0,73467	&	0,26533	&	0,00000	\\
830	&	1,000	&	0,000	&	0,000	&	0,00001	&	0,17697	&	0,00006	&	0,00000	&	0,00000	&	0,73467	&	0,26533	&	0,00000	


																	
\end{pmatrix}

take almost 100ms for rendering. The output is larger than 64kB (it does not fit into the mathlog table). One of the most obvious reasons is that the numbers are handled as individual digits.

FrederikHennecke1 added a comment.Sep 29 2025, 4:32 PM

I created a small benchmarking script for LocalChecker: https://github.com/FrederikHennecke/mediawiki-extensions-Math-testing/tree/master/benchmarks
I used memcached for caching. The results show that the difference between a hit and a miss in the cache is relatively small. As for the input data, I used the reference.json file from the tests. Btw, this is my first time working with the cache, so take these results with a grain of salt.

benchAllMissCore........................I49 - Mo152.840ms (±1.30%)
benchAllHitCore.........................I49 - Mo148.083ms (±2.42%)
Physikerwelt added a comment.Sep 29 2025, 6:24 PM

Interesting, did you verify that something is actually cached?

FrederikHennecke1 added a comment.Sep 30 2025, 2:27 AM

I thought I did but I was mistaken. Seems like I forgot to set one specific cache value and the caching did not work. I fixed the benchmark and now the results are completely different:

benchAllMissCore........................I49 - Mo195.291ms (±2.60%)
benchAllHitCore.........................I49 - Mo18.568ms (±4.47%)
Physikerwelt added a comment.Sep 30 2025, 5:23 AM

Thank you. That seems to better match my expectations. I'll try to run your github on my machine to see if I can match those values. In https://ceur-ws.org/Vol-1785/W48.pdf we show the rendering time over the input length (Fig. 3). I find that a bit more intuitive (but more complicated) than the median.

FrederikHennecke1 added a comment.Oct 11 2025, 1:21 AM
In T388985#11227827, @Physikerwelt wrote:

Thank you. That seems to better match my expectations. I'll try to run your github on my machine to see if I can match those values. In https://ceur-ws.org/Vol-1785/W48.pdf we show the rendering time over the input length (Fig. 3). I find that a bit more intuitive (but more complicated) than the median.

I also added a test for that. You can see the result in the repo. We can see that the runtime for a HIT is roughly the same for the char length, and the runtime for a MISS does grow with input length.

Bugreporter2 updated the task description. (Show Details)Oct 11 2025, 1:38 AM
Bugreporter2 updated the task description. (Show Details)
Physikerwelt added a comment.Nov 12 2025, 7:05 AM

@FrederikHennecke1 It's still significantly faster with caching. So we need to continue with caching.

I think we should start by using href on the elements instead of adding a. For Chrome, we can implement a polyfill.

FrederikHennecke1 added a comment.Dec 27 2025, 8:28 PM

I just tried to find all formulas in enwiki with QIDs with a short python script but I could only find 161 unique formulas. Is this correct? If that is the case I think that we don't necessarily have to rewrite the cache to cache these formulas. I also tested the performance of with / without caching and got the following results:

Miss: Native MML without links

 count: 100000
  rate:   1069.9/s
 total: 93466.28ms
  mean:     0.93ms
   max:    35.16ms
stddev:     0.59ms

Hit: Native MML without links

 count: 100000
  rate:   6268.3/s
 total: 15953.29ms
  mean:     0.16ms
   max:    43.42ms
stddev:     0.74ms

Miss: Native MML with formula links

 count: 100000
  rate:    467.3/s
 total: 213990.63ms
  mean:     2.14ms
   max:    77.15ms
stddev:     1.69ms

Hit: Native MML with formula links

 count: 100000
  rate:    728.6/s
 total: 137240.21ms
  mean:     1.37ms
   max:    84.56ms
stddev:     1.59ms

I only computed the formula $E=mc^2$ 100,000 times for this short benchmark.

In T388985#11365171, @Physikerwelt wrote:

@FrederikHennecke1 It's still significantly faster with caching. So we need to continue with caching.

I think we should start by using href on the elements instead of adding a. For Chrome, we can implement a polyfill.

For this change, we need to update ext.math.popup.js to make the Popup functionality work. I currently have a working prototype I'll upload in the next few days.

Physikerwelt added a comment.Dec 28 2025, 8:29 PM

@FrederikHennecke1 the W3C mathwg discussion continued. Eventually, only mrow/a elements should have links. I suggest we add mrows when needed, and don't switch to a elements right now. As soon as Chrome has native support for links, we rename all mrow elements to a elements (which is a very minimal change with a lot of effect).

Physikerwelt closed subtask T389461: wikitexvc should create a tree instead of a string as Resolved.Dec 28 2025, 8:39 PM
FrederikHennecke1 created subtask T414289: Use mrow elements for linking instead of anchor tags.Jan 11 2026, 9:45 PM
gerritbot added a comment.Tue, Mar 31, 9:59 AM

Change #1128019 abandoned by Physikerwelt:

[mediawiki/extensions/Math@master] Add support for more complex linking in <math/> formulas

https://gerrit.wikimedia.org/r/1128019

Content licensed under Creative Commons Attribution-ShareAlike (CC BY-SA) 4.0 unless otherwise noted; code licensed under GNU General Public License (GPL) 2.0 or later and other open source licenses. By using this site, you agree to the Terms of Use, Privacy Policy, and Code of Conduct. · Wikimedia Foundation · Privacy Policy · Code of Conduct · Terms of Use · Disclaimer · CC-BY-SA · GPL · Credits