| Message ID | 20190707000052.26057-1-eantoranz@gmail.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [v1] builtin/merge.c - cleanup of code in for-cycle that tests strategies | expand |
On Sat, Jul 6, 2019 at 6:00 PM Edmundo Carmona Antoranz <eantoranz@gmail.com> wrote: > @@ -1645,6 +1631,7 @@ int cmd_merge(int argc, const char **argv, const char *prefix) > * auto resolved the merge cleanly. > */ > if (automerge_was_ok) { > + write_tree_trivial(&result_tree); > ret = finish_automerge(head_commit, head_subsumed, > common, remoteheads, > &result_tree, wt_strategy); I realized later that the call to write_tree_trivial could be included in finish_automerge. Will include this change on v2 of the patch (if there's a v2, depending on feedback).
Edmundo Carmona Antoranz <eantoranz@gmail.com> writes: > Previous code was a little convoluted to follow logic > New code is shorter and logic is easier to follow In the body of the proposed commit log message, please finish each sentence with a full-stop. > - Easier to see what happens when merge is successful > and how --no-commit affects result > - Simpler to see that for-cycle will stop when merge_was_ok is set > - Easier to spot what logic will run through best_strategy > - Easier to see that in case of ret being 2, cycle will continue These bullets are all subjective, and do not add any value to what you already said in the second sentence. > - Keep a single break case (when automerge succedes and a revision will > be created) > - Put together closing actions when automerge succedes if a revision > will be created These are facts that readers can see for themselves and agree with. Something like... The cmd_merge() function has a loop that tries different merge strategies in turn, and stops when a strategy gets a clean merge, while keeping the "best" conflicted merge so far. Make the loop easier to follow by moving the code around, ensuring that there is only one "break" in the loop where an automerge succeeds. Also group the actions that are performed after an automerge succeeds together to a single location, outside and after the loop. perhaps? > Signed-off-by: Edmundo Carmona Antoranz <eantoranz@gmail.com> > --- > builtin/merge.c | 51 ++++++++++++++++++------------------------------- > 1 file changed, 19 insertions(+), 32 deletions(-) > > diff --git a/builtin/merge.c b/builtin/merge.c > index 6e99aead46..94f2713bea 100644 > --- a/builtin/merge.c > +++ b/builtin/merge.c > @@ -1586,7 +1586,7 @@ int cmd_merge(int argc, const char **argv, const char *prefix) > save_state(&stash)) > oidclr(&stash); > > - for (i = 0; i < use_strategies_nr; i++) { > + for (i = 0; !merge_was_ok && i < use_strategies_nr; i++) { > int ret; > if (i) { > printf(_("Rewinding the tree to pristine...\n")); > @@ -1604,40 +1604,26 @@ int cmd_merge(int argc, const char **argv, const char *prefix) > ret = try_merge_strategy(use_strategies[i]->name, > common, remoteheads, > head_commit); > - if (!option_commit && !ret) { > - merge_was_ok = 1; > - /* > - * This is necessary here just to avoid writing > - * the tree, but later we will *not* exit with > - * status code 1 because merge_was_ok is set. > - */ > - ret = 1; > - } > - > - if (ret) { > - /* > - * The backend exits with 1 when conflicts are > - * left to be resolved, with 2 when it does not > - * handle the given merge at all. > - */ > - if (ret == 1) { > - int cnt = evaluate_result(); > - > - if (best_cnt <= 0 || cnt <= best_cnt) { > - best_strategy = use_strategies[i]->name; > - best_cnt = cnt; > + /* > + * The backend exits with 1 when conflicts are > + * left to be resolved, with 2 when it does not > + * handle the given merge at all. > + */ > + if (ret < 2) { > + if (!ret) { > + if (option_commit) { > + /* Automerge succeeded. */ > + automerge_was_ok = 1; > + break; > } > + merge_was_ok = 1; > + } > + int cnt = evaluate_result(); This introduces -Wdeclaration-after-statement, doesn't it? Perhaps just declare the variable at the top of the for loop, next to where the local 'ret' is declared? Other than this single glitch, I think the code with this patch does become easier to follow. > + if (best_cnt <= 0 || cnt <= best_cnt) { > + best_strategy = use_strategies[i]->name; > + best_cnt = cnt; > } > - if (merge_was_ok) > - break; > - else > - continue; > } > - > - /* Automerge succeeded. */ > - write_tree_trivial(&result_tree); > - automerge_was_ok = 1; > - break; > } > > /* > @@ -1645,6 +1631,7 @@ int cmd_merge(int argc, const char **argv, const char *prefix) > * auto resolved the merge cleanly. > */ > if (automerge_was_ok) { > + write_tree_trivial(&result_tree); > ret = finish_automerge(head_commit, head_subsumed, > common, remoteheads, > &result_tree, wt_strategy);
On Mon, Jul 8, 2019 at 2:15 PM Junio C Hamano <gitster@pobox.com> wrote: > In the body of the proposed commit log message, please finish each > sentence with a full-stop. Ok > > These bullets are all subjective, and do not add any value to what > you already said in the second sentence. Ok > > These are facts that readers can see for themselves and agree with. > > Something like... > > The cmd_merge() function has a loop that tries different > merge strategies in turn, and stops when a strategy gets a > clean merge, while keeping the "best" conflicted merge so > far. > > Make the loop easier to follow by moving the code around, > ensuring that there is only one "break" in the loop where > an automerge succeeds. Also group the actions that are > performed after an automerge succeeds together to a single > location, outside and after the loop. > > perhaps? I like it. > > + int cnt = evaluate_result(); > This introduces -Wdeclaration-after-statement, doesn't it? It does! Let me take care of it. (Didn't know about DEVELOPER environment variable up until... 5 minutes ago?) > Perhaps just declare the variable at the top of the for loop, next > to where the local 'ret' is declared? Yep, make sense > > Other than this single glitch, I think the code with this patch does > become easier to follow. Great! Let me come back with v2 during the day. Thanks!
diff --git a/builtin/merge.c b/builtin/merge.c index 6e99aead46..94f2713bea 100644 --- a/builtin/merge.c +++ b/builtin/merge.c @@ -1586,7 +1586,7 @@ int cmd_merge(int argc, const char **argv, const char *prefix) save_state(&stash)) oidclr(&stash); - for (i = 0; i < use_strategies_nr; i++) { + for (i = 0; !merge_was_ok && i < use_strategies_nr; i++) { int ret; if (i) { printf(_("Rewinding the tree to pristine...\n")); @@ -1604,40 +1604,26 @@ int cmd_merge(int argc, const char **argv, const char *prefix) ret = try_merge_strategy(use_strategies[i]->name, common, remoteheads, head_commit); - if (!option_commit && !ret) { - merge_was_ok = 1; - /* - * This is necessary here just to avoid writing - * the tree, but later we will *not* exit with - * status code 1 because merge_was_ok is set. - */ - ret = 1; - } - - if (ret) { - /* - * The backend exits with 1 when conflicts are - * left to be resolved, with 2 when it does not - * handle the given merge at all. - */ - if (ret == 1) { - int cnt = evaluate_result(); - - if (best_cnt <= 0 || cnt <= best_cnt) { - best_strategy = use_strategies[i]->name; - best_cnt = cnt; + /* + * The backend exits with 1 when conflicts are + * left to be resolved, with 2 when it does not + * handle the given merge at all. + */ + if (ret < 2) { + if (!ret) { + if (option_commit) { + /* Automerge succeeded. */ + automerge_was_ok = 1; + break; } + merge_was_ok = 1; + } + int cnt = evaluate_result(); + if (best_cnt <= 0 || cnt <= best_cnt) { + best_strategy = use_strategies[i]->name; + best_cnt = cnt; } - if (merge_was_ok) - break; - else - continue; } - - /* Automerge succeeded. */ - write_tree_trivial(&result_tree); - automerge_was_ok = 1; - break; } /* @@ -1645,6 +1631,7 @@ int cmd_merge(int argc, const char **argv, const char *prefix) * auto resolved the merge cleanly. */ if (automerge_was_ok) { + write_tree_trivial(&result_tree); ret = finish_automerge(head_commit, head_subsumed, common, remoteheads, &result_tree, wt_strategy);
Previous code was a little convoluted to follow logic New code is shorter and logic is easier to follow - Easier to see what happens when merge is successful and how --no-commit affects result - Simpler to see that for-cycle will stop when merge_was_ok is set - Easier to spot what logic will run through best_strategy - Easier to see that in case of ret being 2, cycle will continue - Keep a single break case (when automerge succedes and a revision will be created) - Put together closing actions when automerge succedes if a revision will be created Signed-off-by: Edmundo Carmona Antoranz <eantoranz@gmail.com> --- builtin/merge.c | 51 ++++++++++++++++++------------------------------- 1 file changed, 19 insertions(+), 32 deletions(-)