This is an archive of the discontinued Mercurial Phabricator instance.

Differential D4027

revlog: remove micro-optimization for looking up only nullid
ClosedPublic

Authored by martinvonz on Aug 1 2018, 3:01 PM.

Details

Reviewers
None
Group Reviewers
hg-reviewers
Commits
rHG44bbc89ec5e0: revlog: remove micro-optimization for looking up only nullid
Summary

index_find_node() would call nt_find() before initializing the node
tree. nt_find() would then return -2 unless the requested revision was
the null revision. I can't imagine what scenario that is optimizing
for, and doing the initialization earlier makes the code simpler and
easier to follow, so that's what this patch does.

Diff Detail

Repository
rHG Mercurial
Lint
Automatic diff as part of commit; lint not applicable.
Unit
Automatic diff as part of commit; unit tests not applicable.

Event Timeline

martinvonz created this revision.Aug 1 2018, 3:01 PM
Closed by commit rHG44bbc89ec5e0: revlog: remove micro-optimization for looking up only nullid (authored by martinvonz). · Explain WhyAug 3 2018, 9:44 AM
This revision was automatically updated to reflect the committed changes.

Revision Contents
Changeset List

PathPackages
Mmercurial/cext/revlog.c (9 lines)

Diff 9821

mercurial/cext/revlog.c

int hex) int hex)
{ {
int (*getnybble)(const char *, Py_ssize_t) = hex ? hexdigit : nt_level; int (*getnybble)(const char *, Py_ssize_t) = hex ? hexdigit : nt_level;
int level, maxlevel, off; int level, maxlevel, off;
if (nodelen == 20 && node[0] == '\0' && memcmp(node, nullid, 20) == 0) if (nodelen == 20 && node[0] == '\0' && memcmp(node, nullid, 20) == 0)
return -1; return -1;
if (self->nt == NULL)
return -2;
if (hex) if (hex)
maxlevel = nodelen > 40 ? 40 : (int)nodelen; maxlevel = nodelen > 40 ? 40 : (int)nodelen;
else else
maxlevel = nodelen > 20 ? 40 : ((int)nodelen * 2); maxlevel = nodelen > 20 ? 40 : ((int)nodelen * 2);
for (level = off = 0; level < maxlevel; level++) { for (level = off = 0; level < maxlevel; level++) {
int k = getnybble(node, level); int k = getnybble(node, level);
nodetree *n = &self->nt[off]; nodetree *n = &self->nt[off];
* -2: not found (no exception set) * -2: not found (no exception set)
* rest: valid rev * rest: valid rev
*/ */
static int index_find_node(indexObject *self, static int index_find_node(indexObject *self,
const char *node, Py_ssize_t nodelen) const char *node, Py_ssize_t nodelen)
{ {
int rev; int rev;
if (nt_init(self) == -1)
return -3;
self->ntlookups++; self->ntlookups++;
rev = nt_find(self, node, nodelen, 0); rev = nt_find(self, node, nodelen, 0);
if (rev >= -1) if (rev >= -1)
return rev; return rev;
if (nt_init(self) == -1)
return -3;
/* /*
* For the first handful of lookups, we scan the entire index, * For the first handful of lookups, we scan the entire index,
* and cache only the matching nodes. This optimizes for cases * and cache only the matching nodes. This optimizes for cases
* like "hg tip", where only a few nodes are accessed. * like "hg tip", where only a few nodes are accessed.
* *
* After that, we cache every node we visit, using a single * After that, we cache every node we visit, using a single
* scan amortized over multiple lookups. This gives the best * scan amortized over multiple lookups. This gives the best
* bulk performance, e.g. for "hg log". * bulk performance, e.g. for "hg log".