| Message ID | 20211220233459.45739-1-chooglen@google.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | implement branch --recurse-submodules | expand |
Glen Choo <chooglen@google.com> writes: > Submodule branching RFC: > https://lore.kernel.org/git/kl6lv912uvjv.fsf@chooglen-macbookpro.roam.corp.google.com/ > > Original Submodule UX RFC/Discussion: > https://lore.kernel.org/git/YHofmWcIAidkvJiD@google.com/ > > Contributor Summit submodules Notes: > https://lore.kernel.org/git/nycvar.QRO.7.76.6.2110211148060.56@tvgsbejvaqbjf.bet/ > > Submodule UX overhaul updates: > https://lore.kernel.org/git/?q=Submodule+UX+overhaul+update > > This series implements branch --recurse-submodules as laid out in the > Submodule branching RFC (linked above). If there are concerns about the > UX/behavior, I would appreciate feedback on the RFC thread as well :) > > v6 fixes v5's bad rebase; it was based off a merge commit in 'seen' > instead of js/branch-track-inherit. Since v5 is mostly a no-op, I will > also include a range-diff against v4, which is the last version that > anyone except Junio would care about. As mentioned, here is the range-diff against v4 Range-diff against v4: 1: 751e8ae566 < -: ---------- branch: accept multiple upstream branches for tracking 2: 5d1ebe1495 < -: ---------- branch: add flags and config to inherit tracking 3: 0080a1fb35 < -: ---------- config: require lowercase for branch.autosetupmerge 4: dfdbbaaca5 ! 1: 29669c57b4 branch: move --set-upstream-to behavior to dwim_and_setup_tracking() @@ Commit message This refactor is motivated by a desire to add a "dry_run" parameter to create_branch() that will validate whether or not a branch can be - created without actually creating it - this behavior be used in a + created without actually creating it - this behavior will be used in a subsequent commit that adds `git branch --recurse-submodules topic`. Adding "dry_run" is not obvious because create_branch() is also used to @@ branch.c: static void setup_tracking(const char *new_ref, const char *orig_ref, if (tracking.matches > 1) @@ branch.c: static void setup_tracking(const char *new_ref, const char *orig_ref, - tracking.srcs) < 0) + tracking.remote, tracking.srcs) < 0) exit(-1); +cleanup: 5: e22a177cb7 ! 2: ac2532a953 branch: make create_branch() always create a branch @@ Commit message create_branch() was formerly used to set tracking without creating a branch. Since the previous commit replaces this use case with - setup_tracking(), we can simplify create_branch() so that it always - creates a branch. + dwim_and_setup_tracking(), we can simplify create_branch() so that it + always creates a branch. Do this simplification, in particular: 6: 8a895aa401 ! 3: a0ed3fa438 branch: add a dry_run parameter to create_branch() @@ builtin/checkout.c: static void update_refs_for_switch(const struct checkout_opt - opts->track); + opts->track, + 0); - new_branch_info->name = opts->new_branch; - setup_branch_path(new_branch_info); - } + free(new_branch_info->name); + free(new_branch_info->refname); + new_branch_info->name = xstrdup(opts->new_branch); 7: 971c53ec85 = 4: ebded31c96 builtin/branch: clean up action-picking logic in cmd_branch() 8: cd88f3ad92 = 5: 0a7ec6ee75 branch: add --recurse-submodules option for branch creation > This version is based off Josh's js/branch-track-inherit v7. Because that > is not yet in 'seen', I applied those patches onto abe6bb3905 (The first > batch to start the current cycle, 2021-11-29). The base commit can be > found at https://github.com/chooglen/git/tree/js/branch-track-inherit-v7. > I really hope I got right this time, but if not, hopefully the paper > trail is obvious. > > Future work: > * `git branch -d --recurse-submodules` so that users can clean up > extraneous branches. > * `git [checkout | switch] --recurse-submodules` + > submodule.propagateBranches so that users can actually checkout the > branches. > * After [1], it seems clear that --recurse-submodules parsing could > really benefit from some standardization. It's not obvious which > RECURSE_SUBMODULE_* enums are applicable to which commands, and there > is no way to distinguish between an explicit --recurse-submodules from > argv vs submodule.recurse from the config. > > I chose not to use them in this series because their usage is already > inconsistent (grep.c doesn't use them either), and it would be _more_ > confusing to use the enum (handling RECURSE_SUBMODULES_DEFAULT = 1 is > trickier than boolean 0 and 1). > > At this point, I think it would be too noisy to introduce the enum, > but this would be a nice cleanup to do later. > * As documented in branch.c, we create branches using a child process > only because install_branch_config() does not support submodules. > It should be possible to remove the child process once we make the > appropriate changes to config.c. I attempted this in [2] but chose to > punt it because it was too time-consuming at the time. > > Changes since v5: > * Rebase onto v7 of js/branch-track-inherit > (https://lore.kernel.org/git/cover.1639717481.git.steadmon@google.com) > > Changes since v4: > * Rebase correctly onto 'gitster/seen^{/^Merge branch .js/branch-track-inherit.}' > (see base-commit) as suggested in [3] (thanks Junio!) > * These patches were also verified on top of 'next'. > > Changes since v3: > * Split up the old patch 1. Patch 1 had a big diff because it used to > move lines, remove dead code and introduce repo_* functions (thanks > Jonathan!) > ** repo_* functions have been dropped; they added noise and are not > necessary for correctness. > * Use a new, harder-to-misuse function in --set-upstream-to, > dwim_and_setup_tracking(). Now, setup_tracking() never does DWIM and > dwim_and_setup_tracking() always does DWIM. > * Move create_branch() dry_run to its own patch. > * Fix an oversight where submodules in subtrees were ignored. This was > because submodules_of_tree() and tree_entry() didn't recurse into > subtrees. Test this accordingly (thanks Jonathan!). > * cmd_branch() possible actions are more consistently ordered. > * Documentation fixes (thanks Philippe!). > * Additional comments and explanation. > * Drop patch 5 (optional cleanup). > * Rebase onto js/branch-track-inherit v6. > > Changes since v2: > * Rebase onto js/branch-track-inherit. This series should continue to be > the case going forward. > * Patch 1 has a smaller diff because the introduction of > validate_branch_start() no longer changes the function order thanks to a > forward declaration. This artificial forward declaration is removed in a > patch 2 (which can just be squashed into patch 1). > * Optional cleanup: fix questionable exit codes in patch 5. > > Changes since v1: > * Move the functionality of "git branch --dry-run" into "git submodule-helper create-branch --dry-run" > * Add more fields to the submodules_of_tree() struct to reduce the > number of allocations made by the caller. Move this functionality > to patch 3 (formerly patch 4) and drop patch 1. > * Make submodules_of_tree() ignore inactive submodules > * Structure the output of the submodules a bit better by adding prefixes > to the child process' output (instead of inconsistently indenting the > output). > ** I wasn't able to find a good way to interleave stdout/stderr > correctly, so a less-than-desirable workaround was to route the child > process output to stdout/stderr depending on the exit code. > ** Eventually, I would like to structure the output of submodules in a > report, as Ævar suggested. But at this stage, I think that it's > better to spend time getting user feedback on the submodules > branching UX and it'll be easier to standardize the output when we've > implemented more of the UX :) > > [1] https://lore.kernel.org/git/kl6lbl1p9zjf.fsf@chooglen-macbookpro.roam.corp.google.com/ > [2] https://lore.kernel.org/git/kl6lv90ytd4v.fsf@chooglen-macbookpro.roam.corp.google.com/ > [3] https://lore.kernel.org/git/xmqqlf0lz6os.fsf@gitster.g > > Glen Choo (5): > branch: move --set-upstream-to behavior to dwim_and_setup_tracking() > branch: make create_branch() always create a branch > branch: add a dry_run parameter to create_branch() > builtin/branch: clean up action-picking logic in cmd_branch() > branch: add --recurse-submodules option for branch creation > > Documentation/config/advice.txt | 3 + > Documentation/config/submodule.txt | 24 ++- > Documentation/git-branch.txt | 11 +- > advice.c | 1 + > advice.h | 1 + > branch.c | 257 ++++++++++++++++++++----- > branch.h | 57 +++++- > builtin/branch.c | 70 +++++-- > builtin/checkout.c | 3 +- > builtin/submodule--helper.c | 38 ++++ > submodule-config.c | 60 ++++++ > submodule-config.h | 34 ++++ > submodule.c | 11 +- > submodule.h | 3 + > t/t3200-branch.sh | 17 ++ > t/t3207-branch-submodule.sh | 291 +++++++++++++++++++++++++++++ > 16 files changed, 805 insertions(+), 76 deletions(-) > create mode 100755 t/t3207-branch-submodule.sh > > Range-diff against v5: > 1: a9d1108b3e ! 1: 29669c57b4 branch: move --set-upstream-to behavior to dwim_and_setup_tracking() > @@ branch.c: static void setup_tracking(const char *new_ref, const char *orig_ref, > > if (tracking.matches > 1) > @@ branch.c: static void setup_tracking(const char *new_ref, const char *orig_ref, > - tracking.srcs) < 0) > + tracking.remote, tracking.srcs) < 0) > exit(-1); > > +cleanup: > @@ branch.c: N_("\n" > @@ branch.c: void create_branch(struct repository *r, > > if ((commit = lookup_commit_reference(r, &oid)) == NULL) > - die(_("not a valid branch point: '%s'"), start_name); > + die(_("Not a valid branch point: '%s'."), start_name); > - oidcpy(&oid, &commit->object.oid); > + if (out_real_ref) > + *out_real_ref = real_ref ? xstrdup(real_ref) : NULL; > 2: c543c1412a = 2: ac2532a953 branch: make create_branch() always create a branch > 3: dddd434d7a = 3: a0ed3fa438 branch: add a dry_run parameter to create_branch() > 4: 41cca3bd52 = 4: ebded31c96 builtin/branch: clean up action-picking logic in cmd_branch() > 5: 540eeab183 ! 5: 0a7ec6ee75 branch: add --recurse-submodules option for branch creation > @@ t/t3207-branch-submodule.sh (new) > + test_must_fail git rev-parse branch-a && > + > + cat >expected <<-EOF && > -+ submodule ${SQ}sub${SQ}: fatal: a branch named ${SQ}branch-a${SQ} already exists > ++ submodule ${SQ}sub${SQ}: fatal: A branch named ${SQ}branch-a${SQ} already exists. > + fatal: submodule ${SQ}sub${SQ}: cannot create branch ${SQ}branch-a${SQ} > + EOF > + test_cmp expected actual > > base-commit: e1c6879cf24f90693ca108542ec1db970764e1dc > -- > 2.33.GIT
Glen Choo <chooglen@google.com> writes: > v6 fixes v5's bad rebase; it was based off a merge commit in 'seen' > instead of js/branch-track-inherit. Since v5 is mostly a no-op, I will > also include a range-diff against v4, which is the last version that > anyone except Junio would care about. > > This version is based off Josh's js/branch-track-inherit v7. Because that > is not yet in 'seen',... Thanks. js/branch/track-inherit has been using v7 since Fri Dec 17 15:47:47 2021 -0800, but unfortunately that was after the week's integration work and the 'seen' shown to the public probably did not have it. FYI at least since [*1*] was used to create 751363af (branch: add flags and config to inherit tracking, 2021-10-16), the topic js/branch-track-inherit has always been queued on top of f443b226 (Thirteenth batch, 2021-10-14). I try to keep the same base to keep things stable, unless there is a strong enough reason why we should depend on newer base commit. I've wiggled the patches to queue your v6 but haven't pushed the result out yet. Thanks. [Reference] *1* <b9356d9837479914bcf9a265f52afe170be7e2e2.1634445482.git.steadmon@google.com>
Junio C Hamano <gitster@pobox.com> writes: > Glen Choo <chooglen@google.com> writes: > >> v6 fixes v5's bad rebase; it was based off a merge commit in 'seen' >> instead of js/branch-track-inherit. Since v5 is mostly a no-op, I will >> also include a range-diff against v4, which is the last version that >> anyone except Junio would care about. >> >> This version is based off Josh's js/branch-track-inherit v7. Because that >> is not yet in 'seen',... > > Thanks. js/branch/track-inherit has been using v7 since Fri Dec 17 > 15:47:47 2021 -0800, but unfortunately that was after the week's > integration work and the 'seen' shown to the public probably did not > have it. > > FYI at least since [*1*] was used to create 751363af (branch: add > flags and config to inherit tracking, 2021-10-16), the topic > js/branch-track-inherit has always been queued on top of f443b226 > (Thirteenth batch, 2021-10-14). I try to keep the same base to keep > things stable, unless there is a strong enough reason why we should > depend on newer base commit. Ah, I see, so it would make the most sense for me to apply the patches on the the same base as you. I'll keep that in mind in the future.
Submodule branching RFC: https://lore.kernel.org/git/kl6lv912uvjv.fsf@chooglen-macbookpro.roam.corp.google.com/ Original Submodule UX RFC/Discussion: https://lore.kernel.org/git/YHofmWcIAidkvJiD@google.com/ Contributor Summit submodules Notes: https://lore.kernel.org/git/nycvar.QRO.7.76.6.2110211148060.56@tvgsbejvaqbjf.bet/ Submodule UX overhaul updates: https://lore.kernel.org/git/?q=Submodule+UX+overhaul+update This series implements branch --recurse-submodules as laid out in the Submodule branching RFC (linked above). If there are concerns about the UX/behavior, I would appreciate feedback on the RFC thread as well :) v6 fixes v5's bad rebase; it was based off a merge commit in 'seen' instead of js/branch-track-inherit. Since v5 is mostly a no-op, I will also include a range-diff against v4, which is the last version that anyone except Junio would care about. This version is based off Josh's js/branch-track-inherit v7. Because that is not yet in 'seen', I applied those patches onto abe6bb3905 (The first batch to start the current cycle, 2021-11-29). The base commit can be found at https://github.com/chooglen/git/tree/js/branch-track-inherit-v7. I really hope I got right this time, but if not, hopefully the paper trail is obvious. Future work: * `git branch -d --recurse-submodules` so that users can clean up extraneous branches. * `git [checkout | switch] --recurse-submodules` + submodule.propagateBranches so that users can actually checkout the branches. * After [1], it seems clear that --recurse-submodules parsing could really benefit from some standardization. It's not obvious which RECURSE_SUBMODULE_* enums are applicable to which commands, and there is no way to distinguish between an explicit --recurse-submodules from argv vs submodule.recurse from the config. I chose not to use them in this series because their usage is already inconsistent (grep.c doesn't use them either), and it would be _more_ confusing to use the enum (handling RECURSE_SUBMODULES_DEFAULT = 1 is trickier than boolean 0 and 1). At this point, I think it would be too noisy to introduce the enum, but this would be a nice cleanup to do later. * As documented in branch.c, we create branches using a child process only because install_branch_config() does not support submodules. It should be possible to remove the child process once we make the appropriate changes to config.c. I attempted this in [2] but chose to punt it because it was too time-consuming at the time. Changes since v5: * Rebase onto v7 of js/branch-track-inherit (https://lore.kernel.org/git/cover.1639717481.git.steadmon@google.com) Changes since v4: * Rebase correctly onto 'gitster/seen^{/^Merge branch .js/branch-track-inherit.}' (see base-commit) as suggested in [3] (thanks Junio!) * These patches were also verified on top of 'next'. Changes since v3: * Split up the old patch 1. Patch 1 had a big diff because it used to move lines, remove dead code and introduce repo_* functions (thanks Jonathan!) ** repo_* functions have been dropped; they added noise and are not necessary for correctness. * Use a new, harder-to-misuse function in --set-upstream-to, dwim_and_setup_tracking(). Now, setup_tracking() never does DWIM and dwim_and_setup_tracking() always does DWIM. * Move create_branch() dry_run to its own patch. * Fix an oversight where submodules in subtrees were ignored. This was because submodules_of_tree() and tree_entry() didn't recurse into subtrees. Test this accordingly (thanks Jonathan!). * cmd_branch() possible actions are more consistently ordered. * Documentation fixes (thanks Philippe!). * Additional comments and explanation. * Drop patch 5 (optional cleanup). * Rebase onto js/branch-track-inherit v6. Changes since v2: * Rebase onto js/branch-track-inherit. This series should continue to be the case going forward. * Patch 1 has a smaller diff because the introduction of validate_branch_start() no longer changes the function order thanks to a forward declaration. This artificial forward declaration is removed in a patch 2 (which can just be squashed into patch 1). * Optional cleanup: fix questionable exit codes in patch 5. Changes since v1: * Move the functionality of "git branch --dry-run" into "git submodule-helper create-branch --dry-run" * Add more fields to the submodules_of_tree() struct to reduce the number of allocations made by the caller. Move this functionality to patch 3 (formerly patch 4) and drop patch 1. * Make submodules_of_tree() ignore inactive submodules * Structure the output of the submodules a bit better by adding prefixes to the child process' output (instead of inconsistently indenting the output). ** I wasn't able to find a good way to interleave stdout/stderr correctly, so a less-than-desirable workaround was to route the child process output to stdout/stderr depending on the exit code. ** Eventually, I would like to structure the output of submodules in a report, as Ævar suggested. But at this stage, I think that it's better to spend time getting user feedback on the submodules branching UX and it'll be easier to standardize the output when we've implemented more of the UX :) [1] https://lore.kernel.org/git/kl6lbl1p9zjf.fsf@chooglen-macbookpro.roam.corp.google.com/ [2] https://lore.kernel.org/git/kl6lv90ytd4v.fsf@chooglen-macbookpro.roam.corp.google.com/ [3] https://lore.kernel.org/git/xmqqlf0lz6os.fsf@gitster.g Glen Choo (5): branch: move --set-upstream-to behavior to dwim_and_setup_tracking() branch: make create_branch() always create a branch branch: add a dry_run parameter to create_branch() builtin/branch: clean up action-picking logic in cmd_branch() branch: add --recurse-submodules option for branch creation Documentation/config/advice.txt | 3 + Documentation/config/submodule.txt | 24 ++- Documentation/git-branch.txt | 11 +- advice.c | 1 + advice.h | 1 + branch.c | 257 ++++++++++++++++++++----- branch.h | 57 +++++- builtin/branch.c | 70 +++++-- builtin/checkout.c | 3 +- builtin/submodule--helper.c | 38 ++++ submodule-config.c | 60 ++++++ submodule-config.h | 34 ++++ submodule.c | 11 +- submodule.h | 3 + t/t3200-branch.sh | 17 ++ t/t3207-branch-submodule.sh | 291 +++++++++++++++++++++++++++++ 16 files changed, 805 insertions(+), 76 deletions(-) create mode 100755 t/t3207-branch-submodule.sh Range-diff against v5: 1: a9d1108b3e ! 1: 29669c57b4 branch: move --set-upstream-to behavior to dwim_and_setup_tracking() @@ branch.c: static void setup_tracking(const char *new_ref, const char *orig_ref, if (tracking.matches > 1) @@ branch.c: static void setup_tracking(const char *new_ref, const char *orig_ref, - tracking.srcs) < 0) + tracking.remote, tracking.srcs) < 0) exit(-1); +cleanup: @@ branch.c: N_("\n" @@ branch.c: void create_branch(struct repository *r, if ((commit = lookup_commit_reference(r, &oid)) == NULL) - die(_("not a valid branch point: '%s'"), start_name); + die(_("Not a valid branch point: '%s'."), start_name); - oidcpy(&oid, &commit->object.oid); + if (out_real_ref) + *out_real_ref = real_ref ? xstrdup(real_ref) : NULL; 2: c543c1412a = 2: ac2532a953 branch: make create_branch() always create a branch 3: dddd434d7a = 3: a0ed3fa438 branch: add a dry_run parameter to create_branch() 4: 41cca3bd52 = 4: ebded31c96 builtin/branch: clean up action-picking logic in cmd_branch() 5: 540eeab183 ! 5: 0a7ec6ee75 branch: add --recurse-submodules option for branch creation @@ t/t3207-branch-submodule.sh (new) + test_must_fail git rev-parse branch-a && + + cat >expected <<-EOF && -+ submodule ${SQ}sub${SQ}: fatal: a branch named ${SQ}branch-a${SQ} already exists ++ submodule ${SQ}sub${SQ}: fatal: A branch named ${SQ}branch-a${SQ} already exists. + fatal: submodule ${SQ}sub${SQ}: cannot create branch ${SQ}branch-a${SQ} + EOF + test_cmp expected actual base-commit: e1c6879cf24f90693ca108542ec1db970764e1dc