Output Transform Pipeline convert the parsercache HTML string to a DOM. In the process, they currently load data-mw & data-parsoid attributes (which involves converting the string attributes to JSON and then to native classes). This is unnecessary and wasteful since most OTP stages only inspect a small number of nodes and their data-parsoid & data-mw properties.
To improve performance, we need lazy loading of data-parsoid & data-mw properties, especially in post-cache transforms like OTP.
Change #1204733 had a related patch set uploaded (by Subramanya Sastry; author: Subramanya Sastry):
[mediawiki/services/parsoid@master] WIP: lazy data-mw & data-parsoid
Change #1204733 merged by jenkins-bot:
[mediawiki/services/parsoid@master] Implement lazy loading of data-mw and data-parsoid attributes
Change #1228553 had a related patch set uploaded (by Arlolra; author: Arlolra):
[mediawiki/vendor@master] Bump wikimedia/parsoid to 0.23.0-a12
Change #1228553 merged by jenkins-bot:
[mediawiki/vendor@master] Bump wikimedia/parsoid to 0.23.0-a12
data-parsoid lazy loading patches are done and are being deployed. I am going to move this back to 'In Progress' for the WIP patches for lazy loading of data-mw.
Change #1227890 had a related patch set uploaded (by Subramanya Sastry; author: Subramanya Sastry):
[mediawiki/services/parsoid@master] WIP: Make loading truly lazy