Message ID | 20211114173820.687467-1-greenfoo@u92.eu (mailing list archive) |
---|---|
Headers | show |
Series | vimdiff: new layout option + docs | expand |
On Sun, Nov 14, 2021 at 9:38 AM Fernando Ramos <greenfoo@u92.eu> wrote: > > A few weeks ago I presented this RFC [1] where I introduced a new variant of the > vimdiff merge tool ("vimdiff4") that creates three tabs (instead of just one) > that look like this: > > ------------------------------------------ > | <TAB #1> | TAB #2 | TAB #3 | | > ------------------------------------------ > | | | | > | LOCAL | BASE | REMOTE | > | | | | <---- Same information > ------------------------------------------ presented by the > | | "standard" vimdiff > | MERGED | merge tool > | | > ------------------------------------------ > > ------------------------------------------ > | TAB #1 | <TAB #2> | TAB #3 | | > ------------------------------------------ > | | | > | | | > | | | > | BASE | LOCAL | <---- Only differences > | | | between BASE and > | | | LOCAL are shown > | | | > ------------------------------------------ > > ------------------------------------------ > | TAB #1 | TAB #2 | <TAB #3> | | > ------------------------------------------ > | | | > | | | > | | | > | BASE | REMOTE | <---- Only differences > | | | between BASE and > | | | REMOTE are shown > | | | > ------------------------------------------ > > > The motivation behind this was that, for non-trivial merges, the three way diff > presented in the first tab tends to be very confusing and in these cases > indivial diffs between BASE and LOCAL and between BASE and REMOTE are very > useful. > > I have been using a "custom" merge tool for months to achieve this same result > by adding these lines to my .gitconfig file: > > [mergetool "supermerge"] > cmd = vim -f -d -c \"4wincmd w | wincmd J | tabnew | edit $LOCAL | vertical diffsplit $BASE | tabnew | edit $REMOTE | vertical diffsplit $BASE | 2tabprevious\" \"$LOCAL\" \"$BASE\" \"$REMOTE\" \"$MERGED\" > trustExitCode = true > > ...and, because I found this "trick" very useful, I thought it would be a good > idea to add it as a git built-in merge tool (called "vimdiff4" because 1, 2 and > 3 had already been taken) for everyone to use... and that's exactly what the RFC > I published did. > > Now... as you can see in the RFC thread [1], David and Juno suggested that > maybe, instead of creating *yet another vimdiff variant*, we should take this > opportunity to: > > * Come up with a more general way of defining arbitrary vim layouts. > > * Re-implement "vimdiff1", "vimdiff2" and "vimdiff3" using this new mechanism > (after all, the only difference among them is that they present different > layouts to the user) > > * Add documentation to all of this. > > And the result of that work is what I'm presenting today :) > > Some things I would like to mention: > > 1. There are three commits in this patch series: > > - The first one implements the logic to generate new arbitrary layouts and > also re-defines "vimdiff1", "vimdiff2" and "vimdiff3" on top of it. > > - The second one adds documentation. It is probably a good idea to start > reviewing this commit before the first one! > > - The last commit *is not meant to be merged now*. It removes "vimdiff1", > "vimdiff2" and "vimdiff3", which is something that should only be done > after one or two releases with a deprecation notice and only if everyone > agrees to do so :) > > 2. "mergetools/vimdiff" is now a ~800 lines bash script, but most of it is > documentation (which is embedded in the tool itself for easier maintenance) > and unit tests. > I have only tested it with bash, but I've tried not to use any command not > already being used somewhere else, so I expect it to work in the same > places it was working before (however, let me know if there are some shell > compatibility requirements and I'll try to check them). > > 3. Regarding unit tests, "mergetool/vimdiff" contains instructions on how to > run them (just call the script without arguments after making changes, to > make sure you didn't break anything). > Right now it prints "OK" on all test cases (obviously) [2] > > 3. The "git {diff,merge}tool --tool-help" command now also prints the > documentation for each tool (instead of just its name, as before). > You can see an example of the output here ([3] and [4]) > > Finally, let me say that, while I like what this patch series achieves, I would > also *completely* understand if you decide not to merge it due to being a > complex solution to a simple problem that can be solved (as I had been doing up > until today) by just adding three line to one's .gitconfig. > > [mergetool "supermerge"] > cmd = vim -f -d -c ...(custom complex sequence of vim commands)... > trustExitCode = true > > Let me know what you think. > > Thanks. > > References: > > [1] https://lore.kernel.org/git/20211019212020.25385-1-greenfoo@u92.eu/#r > [2] https://pastebin.com/kuQ5pETG > [3] https://pastebin.com/yvLWxeiM > [4] https://pastebin.com/qNc7qymp > > > New in v2: > > * Remove deprecation of vimdiff{1,2,3} (they will remain for backwards > compatibility, but built on top of the new generic layout mechanism) > > * Remove unnecessary "IFS=" > > * Replace DEBUG --> GIT_MERGETOOL_VIMDIFF_DEBUG > > * Stop using "local" (turns out it was not necessary) > > * Stop using bash arrays (use arrays of variables instead) > > * Stop using bash substring expansion (use a new "substring" function > instead). > > * Refactor some of the internal loops to make the code faster. This was needed > because the two previous changes (specially the one where I stop using > "substring expansion") slowed down the script to a point where I had to wait > for a few *seconds* before the "layout" string was resolved (recursion + > bash forks are a recipe for sluggishness). Fortunately we are back to the > low hundreds of milliseconds range. > > * Change markers: > - File to save: * --> @ > - New tab : ; --> + > - Vert split : | --> , > - Horz split : - --> / > > * Rewrite examples to use as many parenthesis as possible (typically zero) > and better explain operators priority. > > * Other fixes to remove problems reported by "shellcheck --shell=sh" (which > checks syntax agains the POSIX shell spec) > > * Rename "index_..." vars to make more obvious what they do. > > * Use "$" inside arithmetic expressions $((...)) > NOTE: "shellcheck" issues a warning stating that "$" is not needed inside > arithmetic expressions. > > * Use "test -n" instead of "! test -z" > > * Use "=" instead of "==" in "test" > > * Use "= 0" instead of "-eq 0" > > * Do not use "eval" nor "--" when copying a file to MERGED > > * Do not use "-o" with grep > > * Use <<-\EOF instead of <<\ENDOFMESSAGE and remove extra indent in "here > docs". > > * Also, let me put here some answers to questions made in the replies to > version v1 of this patch set: > > > What do backticks do in here? > > > > some_var=( > > `# Name` "Rick Deckard" > > `# Age` "Unknown" > > `# Occupation` "Blade runner" > > ) > > The backticks execute the code inside, which is a comment (everything after > a "#" is considered a comment by the shell), thus it does nothing. > > The `# ...` trick can be used to insert inline comments in bash. Another > example: > > $ ls -l `# long format` -h `# human readable` *.txt > > In any case, as you can see in this new revision, because I'm no longer > using bash arrays, this trick to comment each test case is not needed > anymore. > > > Why is "eval" needed here: > > > > eval "$merge_tool_path" \ > > -f "$FINAL_CMD" "$LOCAL" "$BASE" "$REMOTE" "$MERGED" > > Variable "$FINAL_CMD" contains a string that looks like this: > > $ echo $FINAL_CMD > -c "vim cmd 1 | vim cmd 2 | ..." -c "tabfirst" > > We need to call "vim" exactly like that, but if we do this... > > $ vim $FINAL_CMD > > ...then "vim" will be excev'ed with the following arguments: > > 1. -c > 2. "vim > 3. cmd > 4. 1 > ... > > ...instead of the desired scenario: > > 1. -c > 2. cmd 1 | cmd 2 | ... > 3. -c > 4. tabfirst > > Using "eval" fixes this. A shorter example shows how this works: > > $ A="1 \"2 3\"" > > $ ls $A > ls: cannot access '1': No such file or directory > ls: cannot access '"2': No such file or directory > ls: cannot access '3"': No such file or directory > > $ eval ls $A > ls: cannot access '1': No such file or directory > ls: cannot access '2 3': No such file or directory > > * Finally, I think I have addressed all comments to v1 *except for two > things*: > > 1. Moving the documentation to "Documentation/" instead of having it > embedded inside the "mergetools/vimdiff" script. > > 2. Move unit tests to the "test suit" > > For (1) I would like more details. Is the idea not to print any > documentation when running "git mergetool --tool-help" and have all the > details included in "git help mergetool" instead? > Right now "git help mergetool" does not mention any tool specific details > and, instead, redirects the user to "git mergetool --tool-help" (that's why > I originally placed the new documentation there). Okay, that makes sense. Repeating the docs might be fine since I expect that the "--tool-help" help message may end up reading differently from the reference docs, and we do want both. The goal we generally want is to make this information available in the manual pages, etc. which requires writing stuff up in Documentation/. Documentation/git-mergetool--lib.txt might be a useful place for it since that's the code-level central hub for all things related to the mergetools. > In any case, all the doc has been placed on its own commit, so once we > decide how to proceed its just a matter of reverting it. > > For (2) I still need to investigate how this is done and prepare a "v3" if > needed :) Sounds good. One approach might be to source the scriptlet from a new test file in the t/ tree and then running test functions to validate the implementation. You can then drop the snippet at the end that tries to detect if the script is being run directly since the test script will be in charge. We currently have tests in: t/t7610-mergetool.sh t/t7800-difftool.sh Maybe those tests could go into a new t7609-mergetool--lib.sh ?