Minor readme improvements

README.md

@@ -2,24 +2,27 @@
 
 `git-test` is a command-line script for running automated tests against commits in a Git repository. It is especially targeted at developers who like their tests to pass on *every* commit in a branch, not just the branch tip.
 
-The best way to use `git test` is to keep a window open in a second linked worktree of your repository, and as often as you like run
+The best way to use `git test` is to keep a window open in a second linked
+[worktree](#best-practice-use-git-test-in-a-linked-worktree) of your repository,
+and as often as you like run
 
     git test run master..mybranch
 
-`git test` will test the commits in the specified range, reporting any failures. The pass/fail results of running tests are also recorded permanently in your repository as Git "notes" (see `git-notes(1)`).
+`git test` will test the commits in the specified range, reporting any failures.
+The pass/fail results of running tests are also recorded permanently in your
+repository as Git "notes" (see [git-notes(1)](https://www.git-scm.com/docs/git-notes)).
 
 If a commit in the range has already been tested, then by default `git test` reports the old results rather than testing it again. This means that you can run the above command over and over as you work, and `git test` won't repeat tests whose results it already knows. (Of course there are options to allow you to request explicitly that commits be retested.)
 
 The test results are recorded by the *tree* that was tested, not the commit, so old test results remain valid even across some kinds of commit rewriting:
 
-*   If commits are rewritten to change their log messages, authorship, dates, etc., the test results remain valid.
-*   If consecutive commits are squashed, the results remain valid.
-*   If a commit is split into two, only the first (partial) commit needs to be tested.
-*   If some commits deep in a branch are reordered, the test results for commits built on top of the reordered commits often remain valid.
+* If commits are rewritten to change their log messages, authorship, dates, etc., the test results remain valid.
+* If consecutive commits are squashed, the results remain valid.
+* If a commit is split into two, only the first (partial) commit needs to be tested.
+* If some commits deep in a branch are reordered, the test results for commits built on top of the reordered commits often remain valid.
 
 Of course this means that your tests should not depend on things besides the files in the tree. For example, whether your test passes/fails should *not* depend on the current branch name or commit message.
 
-
 ## Usage
 
 ### Defining tests
@@ -65,7 +68,11 @@ You can define multiple tests in a single repository (e.g., cheap vs. expensive
 
 If you have flaky tests that occasionally fail for bogus reasons, you might want to re-run the test against a commit even though `git test` has already recorded a result for that commit. To do so, run `git test run` with the `--force`/`-f` or `--retest` options.
 
-If you want to forget particular old test results without retesting, run `git test run` with the `--forget` option.
+If you want to forget particular old test results without retesting, run
+`git test run` with the `--forget` option, e.g.
+
+    git test run [--test=<name>] --forget commit1
+    git test run [--test=<name>] --forget commit1..commit2
 
 If you want to permanently forget *all* stored results for a particular test (e.g., if something in your environment has changed), run
 
@@ -112,7 +119,6 @@ The last command can be re-run any time; it only does significant work when some
 
 Because linked worktrees share branches and the git configuration with the main repository, test definitions and test results are visible across all worktrees. So you could even run multiple tests at the same time in multiple linked worktrees.
 
-
 ## Installation
 
 Requirements:
@@ -122,43 +128,43 @@ Requirements:
 
 Just put `bin/git-test` somewhere in your `$PATH`, adjusting its first line if necessary to invoke the desired Python interpreter properly in your environment.
 
-
 ## Ideas for future enhancements
 
 Some other features that would be nice:
 
-*   Be more consistent about restoring `HEAD`. `git test run` currently checks out the branch that you started on when it is finished, but only if all of the tests passed. We need some kind of `git test reset` command analogous to `git bisect reset`.
-
-*   `git test bisect`: run `git bisect run` against a range of commits, using a configured test as the command that `bisect` uses to decide whether a commit is good/bad.
+* Be more consistent about restoring `HEAD`. `git test run` currently checks out the branch that you started on when it is finished, but only if all of the tests passed. We need some kind of `git test reset` command analogous to `git bisect reset`.
 
-*   `git test prune`: delete notes for obsolete trees.
+* `git test bisect`: run `git bisect run` against a range of commits, using a configured test as the command that `bisect` uses to decide whether a commit is good/bad.
 
-*   Continuous testing mode, where `git test` watches the repository for changes and re-runs itself automatically whenever the commits it is watching change.
+* `git test prune`: delete notes for obsolete trees.
 
-*   Dependencies between tests; for example:
+* Continuous testing mode, where `git test` watches the repository for changes and re-runs itself automatically whenever the commits it is watching change.
+  For now you can use the [run-continously.sh](run-continously.sh) script.
 
-    *   Provide a way to say "if my `full` test passes, that implies that the `build` test would also pass".
+* Dependencies between tests; for example:
 
-    *   Provide a way to run the `build` test (and record the `build` test's results) as the first step of the `full` test.
+  * Provide a way to say "if my `full` test passes, that implies that the `build` test would also pass".
 
-*   Allow trees to be marked `skip`, if they shouldn't be tested (e.g., due to a known breakage). Perhaps allow the test script to emit a special return code to ask that the commit be marked `skip` (probably following the convention of `git bisect run`).
+  * Provide a way to run the `build` test (and record the `build` test's results) as the first step of the `full` test.
 
-*   Remember return codes and give them back out if the old result is reused.
+* Allow trees to be marked `skip`, if they shouldn't be tested (e.g., due to a known breakage). Perhaps allow the test script to emit a special return code to ask that the commit be marked `skip` (probably following the convention of `git bisect run`).
 
-*   Add a `git test fix <range>`, which starts an interactive rebase, changing the command for the first broken commit from "pick" to "edit".
+* Remember return codes and give them back out if the old result is reused.
 
-*   Support tests that depend on the *commit*, not the *tree*, that they are run against.
+* Add a `git test fix <range>`, which starts an interactive rebase, changing the command for the first broken commit from "pick" to "edit".
 
+* Support tests that depend on the *commit*, not the *tree*, that they are run against.
 
 ## License
 
-`git test` is released under the GPLv2+ license. Pull requests are welcome at the project's GitHub page, https://github.com/mhagger/git-test
-
+`git test` is released under the GPLv2+ license. Pull requests are welcome at the project's GitHub page, <https://github.com/mhagger/git-test>.
 
 ## Caveats and disclaimers
 
 `git test` has pretty good automated tests, but it undoubtedly still has bugs and rough edges. Use it at your own risk.
 
+### Detached head
+
 Please note that when you tell `git test run` to test specified commits, it checks those commits out in your working directory. If the tests fail, it leaves the failing commit checked out *in a detached HEAD state*. This is intentional, so that you can examine the cause of the failure. But it means that if you had changes on your original HEAD that weren't part of any branch, they will now be unreachable.
 
 If you don't know what a detached HEAD state is, please read up on it. Additionally, **it is recommended that you run `git test` in a separate worktree**, which is more convenient anyway (see above for instructions). Note that the `git worktree` command was added in Git release 2.5, so make sure you are using that version of Git or (preferably) newer.
@@ -172,3 +178,19 @@ and
     git test run HEAD
 
 don't change the commit that is checked out, and they won't change your working copy to a detached HEAD state.
+
+### Test result "noise" when viewing history with `--all`
+
+Notice that as a side effect of `git test` saving test results in git notes,
+these notes will become visible "noise" when viewing history with `git log` or
+`gitk` and using the version specifier `--all`. The solution is to tell git
+to ignore those notes with `--exclude` (NB, must be specified *before* `--all`
+in order to have effect).
+
+Writing the full exclude phrase every time will probably be too cumbersome, so
+you most likely want to write wrapper scripts like the following
+
+    #!/bin/sh
+    exec gitk --exclude=refs/notes/* --all "$@" &
+
+or include `--exclude` in your git aliases.

run-continously.sh

@@ -0,0 +1,57 @@
+#!/bin/bash
+
+if [[ $# -lt 1 ]]
+then
+	echo "Usage: $0 [-t name] <commit|range>" 1>&2
+	exit 1
+fi
+
+get_main_worktree() {
+	# Example output:
+
+	# worktree /home/user/src/git-test
+	# HEAD e42d298cff233471c47b50f0f44bb05867109a24
+	# branch refs/heads/main
+	#
+	# worktree /home/user/src/worktree.test.git-test
+	# HEAD 1c0c32641de4a00fa6f5fbd4ddcb478bf5a6ca3f
+	# detached
+
+	# Main worktree is always listed first.
+	git worktree list --porcelain | sed -n '1s/^worktree //p'
+}
+
+wait_for_next_run() {
+	if type -P inotifywait >/dev/null
+	then
+		gitdir=$(git rev-parse --git-dir)
+		case $gitdir
+		in
+			*.git/worktrees*)
+				# Inside a linked worktree so use .git dir for main worktree for monitoring.
+				gitdir=$(cd $(get_main_worktree); git rev-parse --path-format=absolute --git-dir)
+				;;
+			*)
+				;;
+		esac
+		# We need delete_self to pick up changes to HEAD (since it gets renamed
+		# over), and "move" to pick up changes in the refs directories.
+		inotifywait -qq -e delete_self -e move -r "$gitdir/HEAD" "$gitdir/refs"
+	else
+		n=0
+		while [[ $n -lt ${GIT_TEST_CONTINOUS_INTERVAL:-30} ]]
+		do
+			n=$(( n + 1 ))
+			echo -n .
+			sleep 1
+		done
+		echo
+	fi
+}
+
+while true
+do
+	git test run "$@"
+
+	wait_for_next_run
+done