David Jones via llvm-dev
2018-Dec-13 16:31 UTC
[llvm-dev] New LLVM git repository conversion prototype
On Thu, Dec 13, 2018 at 7:18 AM Alex Bradbury via llvm-dev < llvm-dev at lists.llvm.org> wrote:> On Thu, 13 Dec 2018 at 15:07, Hans Wennborg via llvm-dev > <llvm-dev at lists.llvm.org> wrote: > > > > On Thu, Dec 13, 2018 at 3:34 PM James Y Knight <jyknight at google.com> > wrote: > > > > We don't *need* it, no. But it might be useful. > > > > > > The purpose of this tag would not be for people to check out, but > rather to allow "git describe" to return more interesting results when run > on master. > > > > > > See David Jones' email from Nov 16. > > > > Ah, okay. "git describe" is new to me :-) I still think v8.0.0-base is > > a better name though. v8.0.0-dev really sounds like something one > > might check out to get the latest dev version. > >Hmm, I wouldn't worry too much about checking out an old tag by accident... that puts your client into detached head state, and git prints a nice, long warning. (That sort of mistake should be obvious pretty quickly.) v8.0.0-branchpoint might be even clearer, though obviously a little longer.>Whatever the convention we choose, it's going to be painful to change once folks get used to it. (So the wording of the tag is not entirely inconsequential.) My expectation is that the `git describe` aliases will be the typical way to refer to commits in a git world, and aliases like v8.0.0-branchpoint-1234-abcdef seem a bit long to me... when I think about how often I type "rNNNNNN" today, the "-branchpoint" part seems almost obsequious. For a development tag, we should probably try to keep it short, simple, and obvious. After all, we're fundamentally talking about a tag used for development, not releases; something short, like "master-v8" would yield concise, obvious aliases like "master-v8-1234-abcdef01". (Tags on release branches are a fundamentally different matter, and could continue looking exactly like they do today.) FWIW, the conventions I've seen have been inconsistent. One possible consideration: the "-dev" suffix is sometimes used to indicate a dirty workspace, but git describe puts the --dirty suffix after the hash. (So it's unambiguous, in any case.) -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20181213/790fc7dc/attachment.html>
David Greene via llvm-dev
2018-Dec-13 19:25 UTC
[llvm-dev] New LLVM git repository conversion prototype
David Jones via llvm-dev <llvm-dev at lists.llvm.org> writes:> v8.0.0-branchpoint might be even clearer, though obviously a > little longer. > > Whatever the convention we choose, it's going to be painful to change > once folks get used to it. (So the wording of the tag is not entirely > inconsequential.) My expectation is that the `git describe` aliases > will be the typical way to refer to commits in a git world, and > aliases like v8.0.0-branchpoint-1234-abcdef seem a bit long to me... > when I think about how often I type "rNNNNNN" today, the > "-branchpoint" part seems almost obsequious.FWIW we use "split" to mark such commits. So it would be v8.0.0-split. But see below.> For a development tag, we should probably try to keep it short, > simple, and obvious. After all, we're fundamentally talking about a > tag used for development, not releases; something short, like > "master-v8" would yield concise, obvious aliases like > "master-v8-1234-abcdef01". (Tags on release branches are a > fundamentally different matter, and could continue looking exactly > like they do today.)"master" could be confusing for downstream. For example we have a setup where our development happens on "master" and we have a branch called "upstream_master" to pull from upstream. It would be confusing to see a tag called "master" on a branch not named "master." "v8.0.0-split/dev/branchpoint/whatever" could also be confusing to downstream as it could conflict with downstream named tags if their products happen to have version number conflicts. Follwing Duncan's suggestion, would llvm.org-8.0.0-split (or suggest some other suffix) work? I know it's longer but it would be unambiguous. "Duncan P. N. Exon Smith via llvm-dev" <llvm-dev at lists.llvm.org> writes:> I have a mild preference for "llvm.org-8.0.0" (instead of > "llvm-8.0.0"), since llvm.org is unambiguously a point of origin, > whereas "llvm" could just refer to the content. E.g., some downstreams > might package their product with the name "llvm".I have a slight preference for that too, for the same reasons. -David
David Jones via llvm-dev
2018-Dec-13 20:19 UTC
[llvm-dev] New LLVM git repository conversion prototype
On Thu, Dec 13, 2018 at 11:25 AM David Greene <dag at cray.com> wrote:> David Jones via llvm-dev <llvm-dev at lists.llvm.org> writes: > > For a development tag, we should probably try to keep it short, > > simple, and obvious. After all, we're fundamentally talking about a > > tag used for development, not releases; something short, like > > "master-v8" would yield concise, obvious aliases like > > "master-v8-1234-abcdef01". (Tags on release branches are a > > fundamentally different matter, and could continue looking exactly > > like they do today.) > > "master" could be confusing for downstream. For example we have a setup > where our development happens on "master" and we have a branch called > "upstream_master" to pull from upstream. It would be confusing to see a > tag called "master" on a branch not named "master." > > "v8.0.0-split/dev/branchpoint/whatever" could also be confusing to > downstream as it could conflict with downstream named tags if their > products happen to have version number conflicts. Follwing Duncan's > suggestion, would llvm.org-8.0.0-split (or suggest some other suffix) > work? I know it's longer but it would be unambiguous. > >(Just a side note: "split" seems ambiguous to me, in the sense of whether it's before or after the branch is cut: does it point to the start-point of the release branch? Or does it point to the commit that bumps the version number, which is [probably] immediately after? If the former, then `git describe` would attribute the last commit to the wrong release; if the latter, it's really not a "split," since it's only an ancestor of one of the sides. In any case, as long as the name we pick is consistently reachable from master, and not from the release branch; and release branch tags continue not to be reachable from master; then I think we'll be fine.)> "Duncan P. N. Exon Smith via llvm-dev" <llvm-dev at lists.llvm.org> writes: > > > I have a mild preference for "llvm.org-8.0.0" (instead of > > "llvm-8.0.0"), since llvm.org is unambiguously a point of origin, > > whereas "llvm" could just refer to the content. E.g., some downstreams > > might package their product with the name "llvm". > > I have a slight preference for that too, for the same reasons. >I like Duncan's suggestions for release naming, and I could imagine something like: tags/llvm.org/v8.X.Y - development tags/llvm.org/release/v8.0.0-rc1 - release candidate heads/llvm.org/release_80 or heads/llvm.org/release/v8.0 - release branch (the former matches existing branch names) The more I look at it, the more I think the development tag should not contain a valid, complete semver identifier -- those should be reserved for tags pointing to release branches. (The whole impetus of adding dev tag is that the tags on release branches are generally not ancestors of the master HEAD.) Either v8 or v8.0 would satisfy that, since they are missing minor/patch levels; something like v8.X.Y seems even clearer to me. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20181213/d5657e66/attachment.html>
Hans Wennborg via llvm-dev
2018-Dec-13 20:30 UTC
[llvm-dev] New LLVM git repository conversion prototype
On Thu, Dec 13, 2018 at 5:31 PM David Jones <dlj at google.com> wrote:> > On Thu, Dec 13, 2018 at 7:18 AM Alex Bradbury via llvm-dev <llvm-dev at lists.llvm.org> wrote: >> >> On Thu, 13 Dec 2018 at 15:07, Hans Wennborg via llvm-dev >> <llvm-dev at lists.llvm.org> wrote: >> > >> > On Thu, Dec 13, 2018 at 3:34 PM James Y Knight <jyknight at google.com> wrote: >> >> > > We don't *need* it, no. But it might be useful. >> > > >> > > The purpose of this tag would not be for people to check out, but rather to allow "git describe" to return more interesting results when run on master. >> > > >> > > See David Jones' email from Nov 16. >> > >> > Ah, okay. "git describe" is new to me :-) I still think v8.0.0-base is >> > a better name though. v8.0.0-dev really sounds like something one >> > might check out to get the latest dev version. >> > > Hmm, I wouldn't worry too much about checking out an old tag by accident... that puts your client into detached head state, and git prints a nice, long warning. (That sort of mistake should be obvious pretty quickly.)I thought checking them out was the main use for tags. For example, if I want to build version 4.9.0 of gcc, I clone their repo, check out the gcc-4_9_0-release tag, and build. The detached state is fine since I'm not planning to commit there. I might be using git all wrong of course :-)>> v8.0.0-branchpoint might be even clearer, though obviously a little longer.That suggests the tag is where we branched for 8.0.x though, but I think the idea is to put the tag at the version bump, right after the 7.0.x branch point. So I don't think branchpoint is a good name -- it seems we're already thinking it means different things :-)
via llvm-dev
2018-Dec-13 21:59 UTC
[llvm-dev] New LLVM git repository conversion prototype
> -----Original Message----- > From: llvm-dev [mailto:llvm-dev-bounces at lists.llvm.org] On Behalf Of David > Greene via llvm-dev > Sent: Thursday, December 13, 2018 2:26 PM > To: David Jones > Cc: llvm-dev > Subject: Re: [llvm-dev] New LLVM git repository conversion prototype > > David Jones via llvm-dev <llvm-dev at lists.llvm.org> writes: > ... > > For a development tag, we should probably try to keep it short, > > simple, and obvious. After all, we're fundamentally talking about a > > tag used for development, not releases; something short, like > > "master-v8" would yield concise, obvious aliases like > > "master-v8-1234-abcdef01". (Tags on release branches are a > > fundamentally different matter, and could continue looking exactly > > like they do today.) > > "master" could be confusing for downstream. For example we have a setup > where our development happens on "master" and we have a branch called > "upstream_master" to pull from upstream. It would be confusing to see a > tag called "master" on a branch not named "master."Same here. Our "master" is not the upstream "master."> > "v8.0.0-split/dev/branchpoint/whatever" could also be confusing to > downstream as it could conflict with downstream named tags if their > products happen to have version number conflicts. Follwing Duncan's > suggestion, would llvm.org-8.0.0-split (or suggest some other suffix) > work? I know it's longer but it would be unambiguous. > > "Duncan P. N. Exon Smith via llvm-dev" <llvm-dev at lists.llvm.org> writes: > > > I have a mild preference for "llvm.org-8.0.0" (instead of > > "llvm-8.0.0"), since llvm.org is unambiguously a point of origin, > > whereas "llvm" could just refer to the content. E.g., some downstreams > > might package their product with the name "llvm". > > I have a slight preference for that too, for the same reasons.Works for me. We actually put both the upstream and local versions into our release-branch names because we could never keep them straight. --paulr
Alex Bradbury via llvm-dev
2018-Dec-17 16:47 UTC
[llvm-dev] New LLVM git repository conversion prototype
On Thu, 13 Dec 2018 at 20:30, Hans Wennborg <hans at chromium.org> wrote:> > On Thu, Dec 13, 2018 at 7:18 AM Alex Bradbury via llvm-dev <llvm-dev at lists.llvm.org> wrote: > >> v8.0.0-branchpoint might be even clearer, though obviously a little longer. > > That suggests the tag is where we branched for 8.0.x though, but I > think the idea is to put the tag at the version bump, right after the > 7.0.x branch point. So I don't think branchpoint is a good name -- it > seems we're already thinking it means different things :-)Thanks for the catching that, you're right. Best, Alex