bisect: support lazy evaluation for 'skip' option

Summary:
Currently, the 'skip' option is too slow when users want to use it to bisect on
a file or folder like below:

    sl bisect --skip "!( file('path:foo') & file('path:bar') )"

Since it tries to eval the skip revset upfront, which takes a long time for
current `file` implementation. This diff make it lazy evaluated.

Reviewed By: muirdm

Differential Revision: D46643574

fbshipit-source-id: c063fbea2bc0703772e07d5167adc8fbf989a68f

eden/scm/edenscm/commands/__init__.py

@@ -1174,16 +1174,34 @@ the sparse profile from the known %s changeset %s\n"
 
 
 def _update_state(repo, state, rev, good, bad, skip):
-    if rev:
-        nodes = [repo.lookup(i) for i in scmutil.revrange(repo, [rev])]
-    else:
-        nodes = [repo.lookup(".")]
+    def get_nodes(rev):
+        nodes = (
+            [repo.lookup(i) for i in scmutil.revrange(repo, [rev])]
+            if rev
+            else [repo.lookup(".")]
+        )
+        return nodes
+
+    def get_revs(rev):
+        """lazy evaluate `rev` revset expression if it is large.
+
+        This is only used for 'skip' status now.
+        """
+        revs = scmutil.revrange(repo, [rev or "."])
+        fastlen = revs.fastlen()
+        # revset check is slower than nodes check, so avoid revset expr for small revsets.
+        # for example, user might just want to skip a single commit.
+        if fastlen is not None and fastlen < 10:
+            return list(repo.changelog.tonodes(revs))
+        else:
+            return [f"revset:{rev}"]
+
     if good:
-        state["good"] += nodes
+        state["good"] += get_nodes(rev)
     elif bad:
-        state["bad"] += nodes
+        state["bad"] += get_nodes(rev)
     elif skip:
-        state["skip"] += nodes
+        state["skip"] += get_revs(rev)
 
 
 @command(

eden/scm/edenscm/hbisect.py

@@ -37,7 +37,7 @@ def bisect(repo, state):
 
     changelog = repo.changelog
     clparents = changelog.parentrevs
-    skip = set([changelog.rev(n) for n in state["skip"]])
+    skip = _state_to_revs(repo, state, "skip")
 
     def buildancestors(bad, good):
         badrev = min([changelog.rev(n) for n in bad])
@@ -219,6 +219,22 @@ def checkstate(state) -> bool:
         raise error.Abort(_("cannot bisect (no known bad revisions)"))
 
 
+def _state_to_revs(repo, state, kind):
+    items = state[kind]
+    nodes = []
+    revset_exprs = []
+
+    for item in items:
+        if isinstance(item, bytes):
+            nodes.append(item)
+        elif item.startswith("revset:"):
+            revset_exprs.append(item[7:])
+        else:
+            raise error.Abort(_("invalid node: %s, kind: %s") % (item, kind))
+
+    return repo.revs("%ln or %lr", nodes, revset_exprs)
+
+
 def get(repo, status):
     """
     Return a list of revision(s) that match the given status:
@@ -233,7 +249,7 @@ def get(repo, status):
     """
     state = load_state(repo)
     if status in ("good", "bad", "skip", "current"):
-        return list(map(repo.changelog.rev, state[status]))
+        return list(_state_to_revs(repo, state, status))
     else:
         # In the following sets, we do *not* call 'bisect()' with more
         # than one level of recursion, because that can be very, very

eden/scm/tests/test-bisect-skip.t

@@ -18,18 +18,7 @@
 
   $ hg bisect -s "! (file('path:E') or file('path:M'))"
   $ cat .hg/bisect.state
-  skip 426bada5c67598ca65036d57d9e4b64b0c1ce7a0
-  skip 112478962961147124edd43549aedd1a335e44bf
-  skip 26805aba1e600a82e93661149f2313866a221a7b
-  skip f585351a92f85104bff7c284233c338b10eb1df7
-  skip a194cadd16930608adaa649035ad4c16930cbd0f
-  skip 43195508e3bb704c08d24c40375bdd826789dd72
-  skip a31451c3c1debad52cf22ef2aebfc88c75dc899a
-  skip 47eb959e86339c47666b6d1e12c7a9ea534aea1c
-  skip 08eb06eada62c63c386dde447d379684d2a0156d
-  skip fb35f87c67da3431b7514753fc516ec66a60be78
-  skip 652ea04869f63d6b6ce47da40f4fb76e01516a98
-  skip 71d6430ccd823b5187025bfdedc0ac6d8c0f7e34
+  skip revset:! (file('path:E') or file('path:M'))
 
   $ hg bisect -g $A
   $ hg bisect -b $N

eden/scm/tests/test-bisect.t

@@ -246,10 +246,10 @@
   good b5bd63375ab9a290419f2024b7f4ee9ea7ce90a8
   good ed2d2f24b11c368fa8aa0da9f4e1db580abade59
   good 58c80a7c8a4025a94cedaf7b4a4e3124e8909a96
-  skip 9d7d07bc967ca98ad0600c24953fd289ad5fa991
-  skip ce8f0998e922c179e80819d5066fbe46e2998784
-  skip e7fa0811edb063f6319531f0d0a865882138e180
   skip a2e6ea4973e9196ddd3386493b0c214b41fd97d3
+  skip e7fa0811edb063f6319531f0d0a865882138e180
+  skip ce8f0998e922c179e80819d5066fbe46e2998784
+  skip 9d7d07bc967ca98ad0600c24953fd289ad5fa991
 
 # bisect reverse test