Daniel Barkalow [off-list ref] writes:
What, exactly, is -l supposed to do for clone? As far as I can tell, we
automatically do the local magic if we can. Would it be okay to make
"local" default to "if possible", have "-l" mean error if not possible,
and have "--no-local" able to avoid using local magic even if we could use
it?
It used to be that "-l" meant "When it is local, use hardlink if possible
otherwise copy without complaining, as either are cheaper than the pack
piped to unpack." Lack of -l meant no local magic.
Recently lack of -l stopped to mean "no local magic". We still do the
local magic, but we do not do hardlinks and instead do copies. An "-l"
that asks clone across filesystems still falls back to copying but now
gets a warning. "--no-hardlinks" does not have any significance anymore,
as that is what you would get for a local clone without -l.
The way to refuse local magic is to use file://$path/ explicitly; we do
not have --no-local.
On Tue, 4 Mar 2008, Junio C Hamano wrote:
Daniel Barkalow [off-list ref] writes:
quoted
What, exactly, is -l supposed to do for clone? As far as I can tell, we
automatically do the local magic if we can. Would it be okay to make
"local" default to "if possible", have "-l" mean error if not possible,
and have "--no-local" able to avoid using local magic even if we could use
it?
It used to be that "-l" meant "When it is local, use hardlink if possible
otherwise copy without complaining, as either are cheaper than the pack
piped to unpack." Lack of -l meant no local magic.
Recently lack of -l stopped to mean "no local magic". We still do the
local magic, but we do not do hardlinks and instead do copies. An "-l"
that asks clone across filesystems still falls back to copying but now
gets a warning. "--no-hardlinks" does not have any significance anymore,
as that is what you would get for a local clone without -l.
The way to refuse local magic is to use file://$path/ explicitly; we do
not have --no-local.
Ah, okay. I was having a hard time getting the not-using-local-magic case
to be clearly not using local magic, but I think that was confusion on my
part. Also, it looks like we need a test that not using -l is equivalent
to --no-hardlinks.
Would it be okay to additionally block local magic, even with a local
path, if "--no-local" is used? Using our parse-options code means
--no-local is accepted anyway, and it's a bit confusing that it doesn't
block default local magic. (Obviously, this won't break anything, since
old code wouldn't have accepted it.)
-Daniel
*This .sig left intentionally blank*
Junio C Hamano wrote:
Daniel Barkalow [off-list ref] writes:
quoted
What, exactly, is -l supposed to do for clone? As far as I can tell, we
automatically do the local magic if we can. Would it be okay to make
"local" default to "if possible", have "-l" mean error if not possible,
and have "--no-local" able to avoid using local magic even if we could use
it?
It used to be that "-l" meant "When it is local, use hardlink if possible
otherwise copy without complaining, as either are cheaper than the pack
piped to unpack." Lack of -l meant no local magic.
Recently lack of -l stopped to mean "no local magic". We still do the
local magic, but we do not do hardlinks and instead do copies. An "-l"
that asks clone across filesystems still falls back to copying but now
gets a warning. "--no-hardlinks" does not have any significance anymore,
as that is what you would get for a local clone without -l.
Are you sure?
I thought this was changed by 3d5c418f so that hardlinking is done by default.
Lack of -l means "do local magic and use hard links".
Using -l makes git-clone print a warning when hard linking fails.
--no-hardlinks disables hard linking and still does local magic.
-brandon