Hello everyone,
It is possible to passvarious substitution when passing revset to
`repo.revs` (eg: repo.revs("heads(%ld)", [4, 5, 6]).
The current implementation handle this in a quick and dirty way. First a
new revset string is build:
* code: repo.revs("heads(%ld)", [4, 5, 6])
* transformed into (repo.revs("heads(_intlist(4, 5, 6))")
* then revset parser build a syntax tree
* _intlist turns its argument into a baseset
This is clever and simple, it works well for a small number of revision.
However, when the set of revision growth, this becomes ineffective. For
example, discovery might call revset where %ld is replace by hundred of
thousands of revisions. The cost of serializing and serializing them
to/from string become significant.
In addition, the serialization looses any possible optimization from the
source of revisions. For example, if we feed a smartset to `%ld` it
could be nice to keep it a smartset.
We need to improve the current situation by letting the second+ argument
of repo.revs reach the revset logic directly, without a roundtrip
through string.
I was wondering if anyone already thought about the best way to do this
(especially Yuya who tend to have code experiment around revset). If
not, we'll take a stab at it.
Cheers,
_______________________________________________
Mercurial-devel mailing list
Mercurial-devel@mercurial-scm.org
https://www.mercurial-scm.org/mailman/listinfo/mercurial-devel
