| Message ID | 20181031042405.GA5503@sigill.intra.peff.net (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | ls-remote and v2 ref prefixes | expand |
Jeff King <peff@peff.net> writes: > Since b4be74105f (ls-remote: pass ref prefixes when requesting a > remote's refs, 2018-03-15), "ls-remote foo" will pass "refs/heads/foo", > "refs/tags/foo", etc to the transport code in an attempt to let the > other side reduce the size of its advertisement. Jonathan, seeing 2b554353 ("fetch: send "refs/tags/" prefix upon CLI refspecs", 2018-06-05), I am guessing that you are doing the proto v2 work inherited from Brandon? Having to undo this is unfortunate, but I agree with this patch that we need to do this until ref prefix learns to grok wildcards. > Unfortunately this is not correct, as ls-remote patterns do not follow > the usual ref lookup rules, and are in fact tail-matched. So we could > find "refs/heads/foo" or "refs/heads/a/much/deeper/foo" or even > "refs/another/hierarchy/foo". > > Since we can't pass a prefix and there's not yet a v2 extension for > matching wildcards, we must disable this feature to keep the same > behavior as v1. > > Reported-by: Jon Simons <jon@jonsimons.org> > Signed-off-by: Jeff King <peff@peff.net> > --- > builtin/ls-remote.c | 8 -------- > t/t5512-ls-remote.sh | 9 +++++++++ > 2 files changed, 9 insertions(+), 8 deletions(-) > > diff --git a/builtin/ls-remote.c b/builtin/ls-remote.c > index 6a0cdec30d..5faa8459d9 100644 > --- a/builtin/ls-remote.c > +++ b/builtin/ls-remote.c > @@ -88,15 +88,7 @@ int cmd_ls_remote(int argc, const char **argv, const char *prefix) > int i; > pattern = xcalloc(argc, sizeof(const char *)); > for (i = 1; i < argc; i++) { > - const char *glob; > pattern[i - 1] = xstrfmt("*/%s", argv[i]); > - > - glob = strchr(argv[i], '*'); > - if (glob) > - argv_array_pushf(&ref_prefixes, "%.*s", > - (int)(glob - argv[i]), argv[i]); > - else > - expand_ref_prefix(&ref_prefixes, argv[i]); > } > } > > diff --git a/t/t5512-ls-remote.sh b/t/t5512-ls-remote.sh > index bc5703ff9b..ca1b7e5bc1 100755 > --- a/t/t5512-ls-remote.sh > +++ b/t/t5512-ls-remote.sh > @@ -302,4 +302,13 @@ test_expect_success 'ls-remote works outside repository' ' > nongit git ls-remote dst.git > ' > > +test_expect_success 'ls-remote patterns work with all protocol versions' ' > + git for-each-ref --format="%(objectname) %(refname)" \ > + refs/heads/master refs/remotes/origin/master >expect && > + git -c protocol.version=1 ls-remote . master >actual.v1 && > + test_cmp expect actual.v1 && > + git -c protocol.version=2 ls-remote . master >actual.v2 && > + test_cmp expect actual.v2 > +' > + > test_done
> Jeff King <peff@peff.net> writes: > > > Since b4be74105f (ls-remote: pass ref prefixes when requesting a > > remote's refs, 2018-03-15), "ls-remote foo" will pass "refs/heads/foo", > > "refs/tags/foo", etc to the transport code in an attempt to let the > > other side reduce the size of its advertisement. > > Jonathan, seeing 2b554353 ("fetch: send "refs/tags/" prefix upon CLI > refspecs", 2018-06-05), I am guessing that you are doing the proto v2 > work inherited from Brandon? Sorry for the late reply - I had some personal events, but I should be able to look more at Git stuff from now on. Well, it's true that I have been fixing some bugs related to protocol v2. > Having to undo this is unfortunate, but > I agree with this patch that we need to do this until ref prefix learns > to grok wildcards. It is unfortunate, although as far as I can tell, the performance improvements from "git fetch" (which I think is the more frequent command called) remain, so it might not be so bad. I see from your What's Cooking that these patches are to be merged to master, which I agree with.
diff --git a/builtin/ls-remote.c b/builtin/ls-remote.c index 6a0cdec30d..5faa8459d9 100644 --- a/builtin/ls-remote.c +++ b/builtin/ls-remote.c @@ -88,15 +88,7 @@ int cmd_ls_remote(int argc, const char **argv, const char *prefix) int i; pattern = xcalloc(argc, sizeof(const char *)); for (i = 1; i < argc; i++) { - const char *glob; pattern[i - 1] = xstrfmt("*/%s", argv[i]); - - glob = strchr(argv[i], '*'); - if (glob) - argv_array_pushf(&ref_prefixes, "%.*s", - (int)(glob - argv[i]), argv[i]); - else - expand_ref_prefix(&ref_prefixes, argv[i]); } } diff --git a/t/t5512-ls-remote.sh b/t/t5512-ls-remote.sh index bc5703ff9b..ca1b7e5bc1 100755 --- a/t/t5512-ls-remote.sh +++ b/t/t5512-ls-remote.sh @@ -302,4 +302,13 @@ test_expect_success 'ls-remote works outside repository' ' nongit git ls-remote dst.git ' +test_expect_success 'ls-remote patterns work with all protocol versions' ' + git for-each-ref --format="%(objectname) %(refname)" \ + refs/heads/master refs/remotes/origin/master >expect && + git -c protocol.version=1 ls-remote . master >actual.v1 && + test_cmp expect actual.v1 && + git -c protocol.version=2 ls-remote . master >actual.v2 && + test_cmp expect actual.v2 +' + test_done
Since b4be74105f (ls-remote: pass ref prefixes when requesting a remote's refs, 2018-03-15), "ls-remote foo" will pass "refs/heads/foo", "refs/tags/foo", etc to the transport code in an attempt to let the other side reduce the size of its advertisement. Unfortunately this is not correct, as ls-remote patterns do not follow the usual ref lookup rules, and are in fact tail-matched. So we could find "refs/heads/foo" or "refs/heads/a/much/deeper/foo" or even "refs/another/hierarchy/foo". Since we can't pass a prefix and there's not yet a v2 extension for matching wildcards, we must disable this feature to keep the same behavior as v1. Reported-by: Jon Simons <jon@jonsimons.org> Signed-off-by: Jeff King <peff@peff.net> --- builtin/ls-remote.c | 8 -------- t/t5512-ls-remote.sh | 9 +++++++++ 2 files changed, 9 insertions(+), 8 deletions(-)