Skip to contentSkip to conversation

blame: add the ability to ignore commits

PATCH Superseded Barret Rhoden 4 participants 25 messages Apr 10, 2019 – Apr 16, 2019 Documentation

v6 of 10 Superseded Latest is v10 →

lore
Share & tools
lore
#0 lore raw

v6 cover blame: add the ability to ignore commits

This patch set adds the ability to ignore a set of commits and their changes when blaming. This can be used to ignore a commit deemed 'not interesting,' such as reformatting.

The last patch in the series changes the heuristic by which ignored lines are attributed to specific lines in the parent commit. This increases the likelihood of blaming the 'right' commit, where 'right' is subjective. Perhaps we want another algorithm. I'm using a relatively simple one that uses the basics of Michael's fingerprinting code, but he has another algorithm.

v5 -> v6
v5: https://public-inbox.org/git/[email protected]/
- The "guess" heuristic can now look anywhere in the parent file for a matching line, instead of just looking in the parent chunk. The chunks passed to blame_chunk() are smaller than you'd expect: they are just adjacent '-' and '+' sections. Any diff 'context' is a chunk boundary.
- Fixed the parent_len calculation. I had been basing it off of e->num_lines, and treating the blame entry as if it was the target chunk, but the individual blame entries are subsets of the chunk. I just pass the parent chunk info all the way through now.
- Use Michael's newest fingerprinting code, which is a large speedup.
- Made a config option to zero the hash for an ignored line when the heuristic could not find a line in the parent to blame. Previously, this was always 'on'.
- Moved the for loop variable declarations out of the for ().
- Rebased on master.

v4 -> v5
v4: https://public-inbox.org/git/[email protected]/
- Changed the handling of blame_entries from ignored commits so that you can use any algorithm you want to map lines from the diff chunk to different parts of the parent commit.
- fill_origin_blob() optionally can track the offsets of the start of every line, similar to what we do in the scoreboard for the final file. This can be used by the matching algorithm. It has no effect if you are not ignoring commits.
- RFC of a fuzzy/fingerprinting heuristic, based on Michael Platings RFC at https://public-inbox.org/git/[email protected]/
- Made the tests that detect unblamable entries more resilient to different heuristics.
- Fixed a few bugs:
- tests were not grepping the line number from --line-porcelain correctly.
- In the old version, when I passed the "upper" part of the blame entry to the target and marked unblamable, the suspect was incorrectly marked as the parent. The s_lno was also in the parent's address space.

v3 -> v4
v3: https://public-inbox.org/git/[email protected]/
- Cleaned up the tests, especially removing usage of sed -i.
- Squashed the 'tests' commit into the other blame commits. Let me know if you'd like further squashing.

v2 -> v3
v2: https://public-inbox.org/git/[email protected]/
- SHA-1 -> "object name", and fixed other comments
- Changed error string for oidset_parse_file()
- Adjusted existing fsck tests to handle those string changes
- Return hash of all zeros for lines we know we cannot identify
- Allow repeated options for blame.ignoreRevsFile and --ignore-revs-file. An empty file name resets the list. Config options are parsed before the command line options.
- Rebased to master
- Added regression tests

v1 -> v2
v1: https://public-inbox.org/git/[email protected]/
- extracted the skiplist from fsck to avoid duplicating code
- overhauled the interface and options
- split out markIgnoredFiles
- handled merges

Barret Rhoden (6):
Move init_skiplist() outside of fsck
blame: use a helper function in blame_chunk()
blame: add the ability to ignore commits and their changes
blame: add config options to handle output for ignored lines
blame: optionally track line fingerprints during fill_blame_origin()
blame: use a fingerprint heuristic to match ignored lines

 Documentation/blame-options.txt |  18 ++
 Documentation/config/blame.txt  |  16 ++
 Documentation/git-blame.txt     |   1 +
 blame.c                         | 446 ++++++++++++++++++++++++++++----
 blame.h                         |   6 +
 builtin/blame.c                 |  56 ++++
 fsck.c                          |  37 +--
 oidset.c                        |  35 +++
 oidset.h                        |   8 +
 t/t5504-fetch-receive-strict.sh |  14 +-
 t/t8013-blame-ignore-revs.sh    | 202 +++++++++++++++
 11 files changed, 740 insertions(+), 99 deletions(-)
 create mode 100755 t/t8013-blame-ignore-revs.sh
Signature · 1 line
2.21.0.392.gf8f6787159e-goog
#1 lore raw

PATCH v6 1/6 Move init_skiplist() outside of fsck

init_skiplist() took a file consisting of SHA-1s and comments and added the objects to an oidset. This functionality is useful for other commands.

---
 fsck.c                          | 37 +--------------------------------
 oidset.c                        | 35 +++++++++++++++++++++++++++++++
 oidset.h                        |  8 +++++++
 t/t5504-fetch-receive-strict.sh | 14 ++++++-------
 4 files changed, 51 insertions(+), 43 deletions(-)
fsck.c +1 -36
index 2260adb71e7a..d45534ad90f5 100644
@@ -181,41 +181,6 @@ static int fsck_msg_type(enum fsck_msg_id msg_id,
 	return msg_type;
 }
 
-static void init_skiplist(struct fsck_options *options, const char *path)
-{
-	FILE *fp;
-	struct strbuf sb = STRBUF_INIT;
-	struct object_id oid;
-
-	fp = fopen(path, "r");
-	if (!fp)
-		die("Could not open skip list: %s", path);
-	while (!strbuf_getline(&sb, fp)) {
-		const char *p;
-		const char *hash;
-
-		/*
-		 * Allow trailing comments, leading whitespace
-		 * (including before commits), and empty or whitespace
-		 * only lines.
-		 */
-		hash = strchr(sb.buf, '#');
-		if (hash)
-			strbuf_setlen(&sb, hash - sb.buf);
-		strbuf_trim(&sb);
-		if (!sb.len)
-			continue;
-
-		if (parse_oid_hex(sb.buf, &oid, &p) || *p != '\0')
-			die("Invalid SHA-1: %s", sb.buf);
-		oidset_insert(&options->skiplist, &oid);
-	}
-	if (ferror(fp))
-		die_errno("Could not read '%s'", path);
-	fclose(fp);
-	strbuf_release(&sb);
-}
-
 static int parse_msg_type(const char *str)
 {
 	if (!strcmp(str, "error"))
@@ -284,7 +249,7 @@ void fsck_set_msg_types(struct fsck_options *options, const char *values)
 		if (!strcmp(buf, "skiplist")) {
 			if (equal == len)
 				die("skiplist requires a path");
-			init_skiplist(options, buf + equal + 1);
+			oidset_parse_file(&options->skiplist, buf + equal + 1);
 			buf += len + 1;
 			continue;
 		}
oidset.c +35 -0
index fe4eb921df81..878a1b56af1c 100644
@@ -35,3 +35,38 @@ void oidset_clear(struct oidset *set)
 	kh_release_oid(&set->set);
 	oidset_init(set, 0);
 }
+
+void oidset_parse_file(struct oidset *set, const char *path)
+{
+	FILE *fp;
+	struct strbuf sb = STRBUF_INIT;
+	struct object_id oid;
+
+	fp = fopen(path, "r");
+	if (!fp)
+		die("Could not open object name list: %s", path);
+	while (!strbuf_getline(&sb, fp)) {
+		const char *p;
+		const char *name;
+
+		/*
+		 * Allow trailing comments, leading whitespace
+		 * (including before commits), and empty or whitespace
+		 * only lines.
+		 */
+		name = strchr(sb.buf, '#');
+		if (name)
+			strbuf_setlen(&sb, name - sb.buf);
+		strbuf_trim(&sb);
+		if (!sb.len)
+			continue;
+
+		if (parse_oid_hex(sb.buf, &oid, &p) || *p != '\0')
+			die("Invalid object name: %s", sb.buf);
+		oidset_insert(set, &oid);
+	}
+	if (ferror(fp))
+		die_errno("Could not read '%s'", path);
+	fclose(fp);
+	strbuf_release(&sb);
+}
oidset.h +8 -0
index c9d0f6d3cc8b..c4807749df8d 100644
@@ -73,6 +73,14 @@ int oidset_remove(struct oidset *set, const struct object_id *oid);
  */
 void oidset_clear(struct oidset *set);
 
+/**
+ * Add the contents of the file 'path' to an initialized oidset.  Each line is
+ * an unabbreviated object name.  Comments begin with '#', and trailing comments
+ * are allowed.  Leading whitespace and empty or white-space only lines are
+ * ignored.
+ */
+void oidset_parse_file(struct oidset *set, const char *path);
+
 struct oidset_iter {
 	kh_oid_t *set;
 	khiter_t iter;
t/t5504-fetch-receive-strict.sh +7 -7
index 7bc706873c5b..7184f1d07f90 100755
@@ -164,9 +164,9 @@ test_expect_success 'fsck with unsorted skipList' '
 test_expect_success 'fsck with invalid or bogus skipList input' '
 	git -c fsck.skipList=/dev/null -c fsck.missingEmail=ignore fsck &&
 	test_must_fail git -c fsck.skipList=does-not-exist -c fsck.missingEmail=ignore fsck 2>err &&
-	test_i18ngrep "Could not open skip list: does-not-exist" err &&
+	test_i18ngrep "Could not open object name list: does-not-exist" err &&
 	test_must_fail git -c fsck.skipList=.git/config -c fsck.missingEmail=ignore fsck 2>err &&
-	test_i18ngrep "Invalid SHA-1: \[core\]" err
+	test_i18ngrep "Invalid object name: \[core\]" err
 '
 
 test_expect_success 'fsck with other accepted skipList input (comments & empty lines)' '
@@ -193,7 +193,7 @@ test_expect_success 'fsck no garbage output from comments & empty lines errors'
 test_expect_success 'fsck with invalid abbreviated skipList input' '
 	echo $commit | test_copy_bytes 20 >SKIP.abbreviated &&
 	test_must_fail git -c fsck.skipList=SKIP.abbreviated fsck 2>err-abbreviated &&
-	test_i18ngrep "^fatal: Invalid SHA-1: " err-abbreviated
+	test_i18ngrep "^fatal: Invalid object name: " err-abbreviated
 '
 
 test_expect_success 'fsck with exhaustive accepted skipList input (various types of comments etc.)' '
@@ -226,10 +226,10 @@ test_expect_success 'push with receive.fsck.skipList' '
 	test_must_fail git push --porcelain dst bogus &&
 	git --git-dir=dst/.git config receive.fsck.skipList does-not-exist &&
 	test_must_fail git push --porcelain dst bogus 2>err &&
-	test_i18ngrep "Could not open skip list: does-not-exist" err &&
+	test_i18ngrep "Could not open object name list: does-not-exist" err &&
 	git --git-dir=dst/.git config receive.fsck.skipList config &&
 	test_must_fail git push --porcelain dst bogus 2>err &&
-	test_i18ngrep "Invalid SHA-1: \[core\]" err &&
+	test_i18ngrep "Invalid object name: \[core\]" err &&
 
 	git --git-dir=dst/.git config receive.fsck.skipList SKIP &&
 	git push --porcelain dst bogus
@@ -255,10 +255,10 @@ test_expect_success 'fetch with fetch.fsck.skipList' '
 	test_must_fail git --git-dir=dst/.git fetch "file://$(pwd)" $refspec &&
 	git --git-dir=dst/.git config fetch.fsck.skipList does-not-exist &&
 	test_must_fail git --git-dir=dst/.git fetch "file://$(pwd)" $refspec 2>err &&
-	test_i18ngrep "Could not open skip list: does-not-exist" err &&
+	test_i18ngrep "Could not open object name list: does-not-exist" err &&
 	git --git-dir=dst/.git config fetch.fsck.skipList dst/.git/config &&
 	test_must_fail git --git-dir=dst/.git fetch "file://$(pwd)" $refspec 2>err &&
-	test_i18ngrep "Invalid SHA-1: \[core\]" err &&
+	test_i18ngrep "Invalid object name: \[core\]" err &&
 
 	git --git-dir=dst/.git config fetch.fsck.skipList dst/.git/SKIP &&
 	git --git-dir=dst/.git fetch "file://$(pwd)" $refspec
Signature · 1 line
2.21.0.392.gf8f6787159e-goog
#4 lore raw

PATCH v6 2/6 blame: use a helper function in blame_chunk()

In reply to Barret Rhoden

The same code for splitting a blame_entry at a particular line was used twice in blame_chunk(), and I'll use the helper again in an upcoming patch.

---
 blame.c | 44 ++++++++++++++++++++++++++++----------------
 1 file changed, 28 insertions(+), 16 deletions(-)
blame.c +28 -16
index 5c07dec19035..9555a9420836 100644
@@ -839,6 +839,27 @@ static struct blame_entry *reverse_blame(struct blame_entry *head,
 	return tail;
 }
 
+/*
+ * Splits a blame entry into two entries at 'len' lines.  The original 'e'
+ * consists of len lines, i.e. [e->lno, e->lno + len), and the second part,
+ * which is returned, consists of the remainder: [e->lno + len, e->lno +
+ * e->num_lines).  The caller needs to sort out the reference counting for the
+ * new entry's suspect.
+ */
+static struct blame_entry *split_blame_at(struct blame_entry *e, int len,
+					  struct blame_origin *new_suspect)
+{
+	struct blame_entry *n = xcalloc(1, sizeof(struct blame_entry));
+
+	n->suspect = new_suspect;
+	n->lno = e->lno + len;
+	n->s_lno = e->s_lno + len;
+	n->num_lines = e->num_lines - len;
+	e->num_lines = len;
+	e->score = 0;
+	return n;
+}
+
 /*
  * Process one hunk from the patch between the current suspect for
  * blame_entry e and its parent.  This first blames any unfinished
@@ -865,14 +886,9 @@ static void blame_chunk(struct blame_entry ***dstq, struct blame_entry ***srcq,
 		 */
 		if (e->s_lno + e->num_lines > tlno) {
 			/* Move second half to a new record */
-			int len = tlno - e->s_lno;
-			struct blame_entry *n = xcalloc(1, sizeof (struct blame_entry));
-			n->suspect = e->suspect;
-			n->lno = e->lno + len;
-			n->s_lno = e->s_lno + len;
-			n->num_lines = e->num_lines - len;
-			e->num_lines = len;
-			e->score = 0;
+			struct blame_entry *n;
+
+			n = split_blame_at(e, tlno - e->s_lno, e->suspect);
 			/* Push new record to diffp */
 			n->next = diffp;
 			diffp = n;
@@ -919,14 +935,10 @@ static void blame_chunk(struct blame_entry ***dstq, struct blame_entry ***srcq,
 			 * Move second half to a new record to be
 			 * processed by later chunks
 			 */
-			int len = same - e->s_lno;
-			struct blame_entry *n = xcalloc(1, sizeof (struct blame_entry));
-			n->suspect = blame_origin_incref(e->suspect);
-			n->lno = e->lno + len;
-			n->s_lno = e->s_lno + len;
-			n->num_lines = e->num_lines - len;
-			e->num_lines = len;
-			e->score = 0;
+			struct blame_entry *n;
+
+			n = split_blame_at(e, same - e->s_lno,
+					   blame_origin_incref(e->suspect));
 			/* Push new record to samep */
 			n->next = samep;
 			samep = n;
Signature · 1 line
2.21.0.392.gf8f6787159e-goog
#5 lore raw

PATCH v6 3/6 blame: add the ability to ignore commits and their changes

In reply to Barret Rhoden

Commits that make formatting changes or function renames are often not interesting when blaming a file. A user may deem such a commit as 'not interesting' and want to ignore and its changes it when assigning blame.

For example, say a file has the following git history / rev-list:

---O---A---X---B---C---D---Y---E---F

Commits X and Y both touch a particular line, and the other commits do not:

X: "Take a third parameter"
-MyFunc(1, 2);
+MyFunc(1, 2, 3);

Y: "Remove camelcase"
-MyFunc(1, 2, 3);
+my_func(1, 2, 3);

git-blame will blame Y for the change. I'd like to be able to ignore Y:
both the existence of the commit as well as any changes it made. This differs from -S rev-list, which specifies the list of commits to process for the blame. We would still process Y, but just don't let the blame 'stick.'

This patch adds the ability for users to ignore a revision with --ignore-rev=rev, which may be repeated. They can specify a set of files of full object names of revs, e.g. SHA-1 hashes, one per line. A single file may be specified with the blame.ignoreRevFile config option or with --ignore-rev-file=file. Both the config option and the command line option may be repeated multiple times. An empty file name "" will clear the list of revs from previously processed files. Config options are processed before command line options.

For a typical use case, projects will maintain the file containing revisions for commits that perform mass reformatting, and their users have the optional to ignore all of the commits in that file.

Additionally, a user can use the --ignore-rev option for one-off investigation. To go back to the example above, X was a substantive change to the function, but not the change the user is interested in. The user inspected X, but wanted to find the previous change to that line - perhaps a commit that introduced that function call.

To make this work, we can't simply remove all ignored commits from the rev-list. We need to diff the changes introduced by Y so that we can ignore them. We let the blames get passed to Y, just like when processing normally. When Y is the target, we make sure that Y does not *keep* any blames. Any changes that Y is responsible for get passed to its parent. Note we make one pass through all of the scapegoats (parents) to attempt to pass blame normally; we don't know if we *need* to ignore the commit until we've checked all of the parents.

The blame_entry will get passed up the tree until we find a commit that has a diff chunk that affects those lines.

One issue is that the ignored commit *did* make some change, and there is no general solution to finding the line in the parent commit that corresponds to a given line in the ignored commit. That makes it hard to attribute a particular line within an ignored commit's diff correctly.

For example, the parent of an ignored commit has this, say at line 11:

commit-a 11) #include "a.h"
commit-b 12) #include "b.h"

Commit X, which we will ignore, swaps these lines:

commit-X 11) #include "b.h"
commit-X 12) #include "a.h"

We can pass that blame entry to the parent, but line 11 will be attributed to commit A, even though "include b.h" came from commit B. The blame mechanism will be looking at the parent's view of the file at line number 11.

ignore_blame_entry() is set up to allow alternative algorithms for guessing per-line blames. Any line that is not attributed to the parent is marked as 'unblamable', and we output a hash of all zeros.

The existing algorithm is simple: blame each line on the corresponding line in the parent's diff chunk. Any lines beyond that stay with the target.

For example, the parent of an ignored commit has this, say at line 11:

commit-a 11) void new_func_1(void *x, void *y);
commit-b 12) void new_func_2(void *x, void *y);
commit-c 13) some_line_c
commit-d 14) some_line_d

After a commit 'X', we have:

commit-X 11) void new_func_1(void *x,
commit-X 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d

Commit X nets two additionally lines: 13 and 14. The current guess_line_blames() algorithm will not attribute these to the parent, whose diff chunk is only two lines - not four.

When we ignore with the current algorithm, we get:

commit-a 11) void new_func_1(void *x,
commit-b 12) void *y);
00000000 13) void new_func_2(void *x,
00000000 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d

Note that line 12 was blamed on B, though B was the commit for new_func_2(), not new_func_1(). Even when guess_line_blames() finds a line in the parent, it may still be incorrect.

---
 Documentation/blame-options.txt |  14 +++
 Documentation/config/blame.txt  |   7 ++
 Documentation/git-blame.txt     |   1 +
 blame.c                         | 181 ++++++++++++++++++++++++++++++--
 blame.h                         |   3 +
 builtin/blame.c                 |  42 ++++++++
 t/t8013-blame-ignore-revs.sh    | 168 +++++++++++++++++++++++++++++
 7 files changed, 406 insertions(+), 10 deletions(-)
 create mode 100755 t/t8013-blame-ignore-revs.sh
Documentation/blame-options.txt +14 -0
index dc41957afab2..8f155196c6fe 100644
@@ -110,5 +110,19 @@ commit. And the default value is 40. If there are more than one
 `-C` options given, the <num> argument of the last `-C` will
 take effect.
 
+--ignore-rev <rev>::
+	Ignore changes made by the revision when assigning blame, as if the
+	change never happened.  Lines that were changed or added by an ignored
+	commit will be blamed on the previous commit that changed that line or
+	nearby lines.  This option may be specified multiple times to ignore
+	more than one revision.
+
+--ignore-revs-file <file>::
+	Ignore revisions listed in `file`, one unabbreviated object name per line.
+	Whitespace and comments beginning with `#` are ignored.  This option may be
+	repeated, and these files will be processed after any files specified with
+	the `blame.ignoreRevsFile` config option.  An empty file name, `""`, will
+	clear the list of revs from previously processed files.
+
 -h::
 	Show help message.
Documentation/config/blame.txt +7 -0
index 67b5c1d1e02a..4da2788f306d 100644
@@ -19,3 +19,10 @@ blame.showEmail::
 blame.showRoot::
 	Do not treat root commits as boundaries in linkgit:git-blame[1].
 	This option defaults to false.
+
+blame.ignoreRevsFile::
+	Ignore revisions listed in the file, one unabbreviated object name per
+	line, in linkgit:git-blame[1].  Whitespace and comments beginning with
+	`#` are ignored.  This option may be repeated multiple times.  Empty
+	file names will reset the list of ignored revisions.  This option will
+	be handled before the command line option `--ignore-revs-file`.
Documentation/git-blame.txt +1 -0
index 16323eb80e31..7e8154199635 100644
@@ -10,6 +10,7 @@ SYNOPSIS
 [verse]
 'git blame' [-c] [-b] [-l] [--root] [-t] [-f] [-n] [-s] [-e] [-p] [-w] [--incremental]
 	    [-L <range>] [-S <revs-file>] [-M] [-C] [-C] [-C] [--since=<date>]
+	    [--ignore-rev <rev>] [--ignore-revs-file <file>]
 	    [--progress] [--abbrev=<n>] [<rev> | --contents <file> | --reverse <rev>..<rev>]
 	    [--] <file>
 
blame.c +171 -10

266 lines not on this page · raw message

blame.h +3 -0
index be3a895043e0..91664913d7c4 100644
@@ -92,6 +92,7 @@ struct blame_entry {
 	 * scanning the lines over and over.
 	 */
 	unsigned score;
+	int unblamable;
 };
 
 /*
@@ -117,6 +118,8 @@ struct blame_scoreboard {
 	/* linked list of blames */
 	struct blame_entry *ent;
 
+	struct oidset ignore_list;
+
 	/* look-up a line in the final buffer */
 	int num_lines;
 	int *lineno;
builtin/blame.c +42 -0

90 lines not on this page · raw message

t/t8013-blame-ignore-revs.sh new file +168 -0

168 lines not on this page · raw message

Signature · 1 line
2.21.0.392.gf8f6787159e-goog
#10 lore raw

PATCH v6 4/6 blame: add config options to handle output for ignored lines

In reply to Barret Rhoden

When ignoring commits, the commit that is blamed might not be responsible for the change. Users might want to know when a particular line has a potentially inaccurate blame. Furthermore, they might never want to see the object hash of an ignored commit.

This patch adds two config options to control the output behavior.

The first option can identify ignored lines by specifying blame.markIgnoredFiles. When this option is set, each blame line is marked with an '*'.

For example:

	278b6158d6fdb (Barret Rhoden  2016-04-11 13:57:54 -0400 26)
appears as:
	*278b6158d6fd (Barret Rhoden  2016-04-11 13:57:54 -0400 26)

where the '*' is placed before the commit, and the hash has one fewer characters.

Sometimes we are unable to even guess at what commit touched a line. These lines are 'unblamable.' The second option, blame.maskIgnoredUnblamables, will zero the hash of any unblamable line.

For example, say we ignore e5e8d36d04cbe:

	e5e8d36d04cbe (Barret Rhoden  2016-04-11 13:57:54 -0400 26)
appears as:
	0000000000000 (Barret Rhoden  2016-04-11 13:57:54 -0400 26)
---
 Documentation/blame-options.txt |  6 +++++-
 Documentation/config/blame.txt  |  9 +++++++++
 blame.c                         |  4 ++++
 blame.h                         |  1 +
 builtin/blame.c                 | 18 +++++++++++++++--
 t/t8013-blame-ignore-revs.sh    | 34 +++++++++++++++++++++++++++++++++
 6 files changed, 69 insertions(+), 3 deletions(-)
Documentation/blame-options.txt +5 -1
index 8f155196c6fe..e7b8b5e4b87b 100644
@@ -115,7 +115,11 @@ take effect.
 	change never happened.  Lines that were changed or added by an ignored
 	commit will be blamed on the previous commit that changed that line or
 	nearby lines.  This option may be specified multiple times to ignore
-	more than one revision.
+	more than one revision.  If the `blame.markIgnoredLines` config option
+	is set, then lines that were changed by an ignored commit will be
+	marked with a `*` in the blame output.  If the
+	`blame.maskIgnoredUnblamables` config option is set, then those lines that
+	we could not attribute to another revision are outputted as all zeros.
 
 --ignore-revs-file <file>::
 	Ignore revisions listed in `file`, one unabbreviated object name per line.
Documentation/config/blame.txt +9 -0
index 4da2788f306d..bb6674227da1 100644
@@ -26,3 +26,12 @@ blame.ignoreRevsFile::
 	`#` are ignored.  This option may be repeated multiple times.  Empty
 	file names will reset the list of ignored revisions.  This option will
 	be handled before the command line option `--ignore-revs-file`.
+
+blame.maskIgnoredUnblamables::
+	Output an object hash of all zeros for lines that were changed by an ignored
+	revision and that we could not attribute to another revision in the output
+	of linkgit:git-blame[1].
+
+blame.markIgnoredLines::
+	Mark lines that were changed by an ignored revision with a '*' in the
+	output of linkgit:git-blame[1].
blame.c +4 -0
index 0bbb86ad5985..a98ae00e2cfc 100644
@@ -481,6 +481,7 @@ void blame_coalesce(struct blame_scoreboard *sb)
 	for (ent = sb->ent; ent && (next = ent->next); ent = next) {
 		if (ent->suspect == next->suspect &&
 		    ent->s_lno + ent->num_lines == next->s_lno &&
+		    ent->ignored == next->ignored &&
 		    ent->unblamable == next->unblamable) {
 			ent->num_lines += next->num_lines;
 			ent->next = next->next;
@@ -733,6 +734,7 @@ static void split_overlap(struct blame_entry *split,
 	int chunk_end_lno;
 	memset(split, 0, sizeof(struct blame_entry [3]));
 
+	split[0].ignored = split[1].ignored = split[2].ignored = e->ignored;
 	split[0].unblamable = e->unblamable;
 	split[1].unblamable = e->unblamable;
 	split[2].unblamable = e->unblamable;
@@ -857,6 +859,7 @@ static struct blame_entry *split_blame_at(struct blame_entry *e, int len,
 	struct blame_entry *n = xcalloc(1, sizeof(struct blame_entry));
 
 	n->suspect = new_suspect;
+	n->ignored = e->ignored;
 	n->unblamable = e->unblamable;
 	n->lno = e->lno + len;
 	n->s_lno = e->s_lno + len;
@@ -922,6 +925,7 @@ static void ignore_blame_entry(struct blame_entry *e,
 	struct blame_line_tracker *line_blames;
 	int entry_len, nr_lines, i;
 
+	e->ignored = 1;
 	line_blames = xcalloc(sizeof(struct blame_line_tracker),
 			      e->num_lines);
 	guess_line_blames(e, parent, target, offset, parent_slno, parent_len,
blame.h +1 -0
index 91664913d7c4..53df8b4c5b3f 100644
@@ -92,6 +92,7 @@ struct blame_entry {
 	 * scanning the lines over and over.
 	 */
 	unsigned score;
+	int ignored;
 	int unblamable;
 };
 
builtin/blame.c +16 -2
index b48842d8459b..c10a6a802240 100644
@@ -53,6 +53,8 @@ static int show_progress;
 static char repeated_meta_color[COLOR_MAXLEN];
 static int coloring_mode;
 static struct string_list ignore_revs_file_list = STRING_LIST_INIT_NODUP;
+static int mask_ignored_unblamables;
+static int mark_ignored_lines;
 
 static struct date_mode blame_date_mode = { DATE_ISO8601 };
 static size_t blame_date_width;
@@ -347,7 +349,7 @@ static void emit_porcelain(struct blame_scoreboard *sb, struct blame_entry *ent,
 	char hex[GIT_MAX_HEXSZ + 1];
 
 	oid_to_hex_r(hex, &suspect->commit->object.oid);
-	if (ent->unblamable)
+	if (mask_ignored_unblamables && ent->unblamable)
 		memset(hex, '0', strlen(hex));
 	printf("%s %d %d %d\n",
 	       hex,
@@ -482,7 +484,11 @@ static void emit_other(struct blame_scoreboard *sb, struct blame_entry *ent, int
 			}
 		}
 
-		if (ent->unblamable)
+		if (mark_ignored_lines && ent->ignored) {
+			length--;
+			putchar('*');
+		}
+		if (mask_ignored_unblamables && ent->unblamable)
 			memset(hex, '0', length);
 		printf("%.*s", length, hex);
 		if (opt & OUTPUT_ANNOTATE_COMPAT) {
@@ -710,6 +716,14 @@ static int git_blame_config(const char *var, const char *value, void *cb)
 		string_list_insert(&ignore_revs_file_list, str);
 		return 0;
 	}
+	if (!strcmp(var, "blame.maskignoredunblamables")) {
+		mask_ignored_unblamables = git_config_bool(var, value);
+		return 0;
+	}
+	if (!strcmp(var, "blame.markignoredlines")) {
+		mark_ignored_lines = git_config_bool(var, value);
+		return 0;
+	}
 	if (!strcmp(var, "color.blame.repeatedlines")) {
 		if (color_parse_mem(value, strlen(value), repeated_meta_color))
 			warning(_("invalid color '%s' in color.blame.repeatedLines"),
t/t8013-blame-ignore-revs.sh +34 -0
index df4993f98682..cc049a390b0d 100755
@@ -53,6 +53,7 @@ test_expect_success ignore_rev_changing_lines '
 # to have nothing in common with "line-one" or "line-two", to keep any
 # heuristics from matching them with any lines in the parent.
 test_expect_success ignore_rev_adding_unblamable_lines '
+	git config --add blame.maskIgnoredUnblamables true &&
 	test_write_lines line-one-change line-two-changed y3 y4 >file &&
 	git add file &&
 	test_tick &&
@@ -123,6 +124,39 @@ test_expect_success bad_files_and_revs '
 	test_i18ngrep "Invalid object name: NOREV" err
 	'
 
+# Commit Z will touch the first two lines.  Y touched all four.
+# 	A--B--X--Y--Z
+# The blame output when ignoring Z should be:
+# ^Y ... 1)
+# ^Y ... 2)
+# Y  ... 3)
+# Y  ... 4)
+# We're checking only the first character
+test_expect_success mark_ignored_lines '
+	git config --add blame.markIgnoredLines true &&
+
+	test_write_lines line-one-Z line-two-Z y3 y4 >file &&
+	git add file &&
+	test_tick &&
+	git commit -m Z &&
+	git tag Z &&
+
+	git blame --ignore-rev Z file >blame_raw &&
+	echo "*" >expect &&
+
+	sed -n "1p" blame_raw | cut -c1 >actual &&
+	test_cmp expect actual &&
+
+	sed -n "2p" blame_raw | cut -c1 >actual &&
+	test_cmp expect actual &&
+
+	sed -n "3p" blame_raw | cut -c1 >actual &&
+	! test_cmp expect actual &&
+
+	sed -n "4p" blame_raw | cut -c1 >actual &&
+	! test_cmp expect actual
+	'
+
 # Resetting the repo and creating:
 #
 # A--B--M
Signature · 1 line
2.21.0.392.gf8f6787159e-goog
#16 lore raw

PATCH v6 5/6 blame: optionally track line fingerprints during fill_blame_origin()

In reply to Barret Rhoden

fill_blame_origin() is a convenient place to store data that we will use throughout the lifetime of a blame_origin. Some heuristics for ignoring commits during a blame session can make use of this storage. In particular, we will calculate a fingerprint for each line of a file for blame_origins involved in an ignored commit.

In this commit, we only calculate the line_starts, reusing the existing code from the scoreboard's line_starts. In an upcoming commit, we will actually compute the fingerprints.

This feature will be used when we attempt to pass blame entries to parents when we "ignore" a commit. Most uses of fill_blame_origin() will not require this feature, hence the flag parameter. Multiple calls to fill_blame_origin() are idempotent, and any of them can request the creation of the fingerprints structure.

---
 blame.c | 95 +++++++++++++++++++++++++++++++++++++++------------------
 blame.h |  2 ++
 2 files changed, 67 insertions(+), 30 deletions(-)
blame.c +65 -30
index a98ae00e2cfc..a42dff80b1a5 100644
@@ -311,12 +311,63 @@ static int diff_hunks(mmfile_t *file_a, mmfile_t *file_b,
 	return xdi_diff(file_a, file_b, &xpp, &xecfg, &ecb);
 }
 
+static const char *get_next_line(const char *start, const char *end)
+{
+	const char *nl = memchr(start, '\n', end - start);
+
+	return nl ? nl + 1 : end;
+}
+
+static int find_line_starts(int **line_starts, const char *buf,
+			    unsigned long len)
+{
+	const char *end = buf + len;
+	const char *p;
+	int *lineno;
+	int num = 0;
+
+	for (p = buf; p < end; p = get_next_line(p, end))
+		num++;
+
+	ALLOC_ARRAY(*line_starts, num + 1);
+	lineno = *line_starts;
+
+	for (p = buf; p < end; p = get_next_line(p, end))
+		*lineno++ = p - buf;
+
+	*lineno = len;
+
+	return num;
+}
+
+static void fill_origin_fingerprints(struct blame_origin *o, mmfile_t *file)
+{
+	int *line_starts;
+
+	if (o->fingerprints)
+		return;
+	o->num_lines = find_line_starts(&line_starts, o->file.ptr,
+					o->file.size);
+	/* TODO: Will fill in fingerprints in a future commit */
+	o->fingerprints = xcalloc(sizeof(struct fingerprint), o->num_lines);
+	free(line_starts);
+}
+
+static void drop_origin_fingerprints(struct blame_origin *o)
+{
+	if (o->fingerprints) {
+		o->num_lines = 0;
+		FREE_AND_NULL(o->fingerprints);
+	}
+}
+
 /*
  * Given an origin, prepare mmfile_t structure to be used by the
  * diff machinery
  */
 static void fill_origin_blob(struct diff_options *opt,
-			     struct blame_origin *o, mmfile_t *file, int *num_read_blob)
+			     struct blame_origin *o, mmfile_t *file,
+			     int *num_read_blob, int fill_fingerprints)
 {
 	if (!o->file.ptr) {
 		enum object_type type;
@@ -340,11 +391,14 @@ static void fill_origin_blob(struct diff_options *opt,
 	}
 	else
 		*file = o->file;
+	if (fill_fingerprints)
+		fill_origin_fingerprints(o, file);
 }
 
 static void drop_origin_blob(struct blame_origin *o)
 {
 	FREE_AND_NULL(o->file.ptr);
+	drop_origin_fingerprints(o);
 }
 
 /*
@@ -1136,8 +1190,10 @@ static void pass_blame_to_parent(struct blame_scoreboard *sb,
 	d.ignore_diffs = ignore_diffs;
 	d.dstq = &newdest; d.srcq = &target->suspects;
 
-	fill_origin_blob(&sb->revs->diffopt, parent, &file_p, &sb->num_read_blob);
-	fill_origin_blob(&sb->revs->diffopt, target, &file_o, &sb->num_read_blob);
+	fill_origin_blob(&sb->revs->diffopt, parent, &file_p,
+			 &sb->num_read_blob, ignore_diffs);
+	fill_origin_blob(&sb->revs->diffopt, target, &file_o,
+			 &sb->num_read_blob, ignore_diffs);
 	sb->num_get_patch++;
 
 	if (diff_hunks(&file_p, &file_o, blame_chunk_cb, &d, sb->xdl_opts))
@@ -1348,7 +1404,8 @@ static void find_move_in_parent(struct blame_scoreboard *sb,
 	if (!unblamed)
 		return; /* nothing remains for this target */
 
-	fill_origin_blob(&sb->revs->diffopt, parent, &file_p, &sb->num_read_blob);
+	fill_origin_blob(&sb->revs->diffopt, parent, &file_p,
+			 &sb->num_read_blob, 0);
 	if (!file_p.ptr)
 		return;
 
@@ -1477,7 +1534,8 @@ static void find_copy_in_parent(struct blame_scoreboard *sb,
 			norigin = get_origin(parent, p->one->path);
 			oidcpy(&norigin->blob_oid, &p->one->oid);
 			norigin->mode = p->one->mode;
-			fill_origin_blob(&sb->revs->diffopt, norigin, &file_p, &sb->num_read_blob);
+			fill_origin_blob(&sb->revs->diffopt, norigin, &file_p,
+					 &sb->num_read_blob, 0);
 			if (!file_p.ptr)
 				continue;
 
@@ -1816,37 +1874,14 @@ void assign_blame(struct blame_scoreboard *sb, int opt)
 	}
 }
 
-static const char *get_next_line(const char *start, const char *end)
-{
-	const char *nl = memchr(start, '\n', end - start);
-	return nl ? nl + 1 : end;
-}
-
 /*
  * To allow quick access to the contents of nth line in the
  * final image, prepare an index in the scoreboard.
  */
 static int prepare_lines(struct blame_scoreboard *sb)
 {
-	const char *buf = sb->final_buf;
-	unsigned long len = sb->final_buf_size;
-	const char *end = buf + len;
-	const char *p;
-	int *lineno;
-	int num = 0;
-
-	for (p = buf; p < end; p = get_next_line(p, end))
-		num++;
-
-	ALLOC_ARRAY(sb->lineno, num + 1);
-	lineno = sb->lineno;
-
-	for (p = buf; p < end; p = get_next_line(p, end))
-		*lineno++ = p - buf;
-
-	*lineno = len;
-
-	sb->num_lines = num;
+	sb->num_lines = find_line_starts(&sb->lineno, sb->final_buf,
+					 sb->final_buf_size);
 	return sb->num_lines;
 }
 
blame.h +2 -0
index 53df8b4c5b3f..5dd877bb78fc 100644
@@ -51,6 +51,8 @@ struct blame_origin {
 	 */
 	struct blame_entry *suspects;
 	mmfile_t file;
+	int num_lines;
+	void *fingerprints;
 	struct object_id blob_oid;
 	unsigned mode;
 	/* guilty gets set when shipping any suspects to the final
Signature · 1 line
2.21.0.392.gf8f6787159e-goog
#17 lore raw

PATCH v6 6/6 blame: use a fingerprint heuristic to match ignored lines

In reply to Barret Rhoden

This replaces the heuristic used to identify lines from ignored commits with one that finds likely candidate lines in the parent's version of the file.

The old heuristic simply assigned lines in the target to the same line number (plus offset) in the parent. The new function uses a fingerprinting algorithm to detect similarity between lines.

The fingerprint code and the idea to use them for blame came from Michael Platings <[email protected]>.

For each line changed in the target, i.e. in a blame_entry touched by a target's diff, guess_line_blames() finds the best line in the parent, above a magic threshold. Ties are broken by proximity of the parent line number to the target's line.

We actually make two passes. The first pass checks in the diff chunk associated with the blame entry - specifically from blame_chunk(). Often times, those diff chunks are small; any 'context' in a normal diff chunk is broken up into multiple calls to blame_chunk(). We make a second pass over the entire parent, with a slightly higher threshold.

Here's an example of the difference the fingerprinting makes. Consider a file with four commits:

	commit-a 11) void new_func_1(void *x, void *y);
	commit-b 12) void new_func_2(void *x, void *y);
	commit-c 13) some_line_c
	commit-d 14) some_line_d

After a commit 'X', we have:

	commit-X 11) void new_func_1(void *x,
	commit-X 12)                 void *y);
	commit-X 13) void new_func_2(void *x,
	commit-X 14)                 void *y);
	commit-c 15) some_line_c
	commit-d 16) some_line_d

When we blame-ignored with the old algorithm, we get:

	commit-a 11) void new_func_1(void *x,
	commit-b 12)                 void *y);
	00000000 13) void new_func_2(void *x,
	00000000 14)                 void *y);
	commit-c 15) some_line_c
	commit-d 16) some_line_d

Where commit-b is blamed for 12 instead of 13. With the fingerprint algorithm, we get:

	commit-a 11) void new_func_1(void *x,
	commit-b 12)                 void *y);
	commit-b 13) void new_func_2(void *x,
	commit-b 14)                 void *y);
	commit-c 15) some_line_c
	commit-d 16) some_line_d

Note both lines 12 and 14 are given to commit b. Their match is above the FINGERPRINT_CHUNK_THRESHOLD, and they tied. Specifically, parent lines 11 and 12 both match these lines. The algorithm chose parent line 12, since that was closest to the target line numbers of 12 and 14.

If we increase the threshold, say to 10, those two lines won't match, and will be treated as 'unblamable.'

For an example of scanning the entire parent for a match, consider:

	commit-a 30) #include <sys/header_a.h>
	commit-b 31) #include <header_b.h>
	commit-c 32) #include <header_c.h>

Then commit X alphabetizes them:

	commit-X 30) #include <header_b.h>
	commit-X 31) #include <header_c.h>
	commit-X 32) #include <sys/header_a.h>

If we just check the parent's chunk (i.e. the first pass), we'd get:

	commit-b 30) #include <header_b.h>
	commit-c 31) #include <header_c.h>
	00000000 32) #include <sys/header_a.h>

That's because commit X consists of two chunks: one chunk is removing sys/header_a.h, then some context, and the second chunk is adding sys/header_a.h.

If we scan the entire parent file, we get:

	commit-b 30) #include <header_b.h>
	commit-c 31) #include <header_c.h>
	commit-a 32) #include <sys/header_a.h>
---
 blame.c | 140 ++++++++++++++++++++++++++++++++++++++++++++++++++++----
 1 file changed, 131 insertions(+), 9 deletions(-)
blame.c +131 -9
index a42dff80b1a5..da2d664b38af 100644
@@ -340,6 +340,84 @@ static int find_line_starts(int **line_starts, const char *buf,
 	return num;
 }
 
+struct fingerprint {
+	struct hashmap map;
+	struct hashmap_entry *entries;
+};
+
+static void get_fingerprint(struct fingerprint *result,
+			    const char *line_begin,
+			    const char *line_end)
+{
+	unsigned int hash;
+	char c0, c1;
+	const char *p;
+	int map_entry_count = line_end - line_begin - 1;
+	struct hashmap_entry *entry = xcalloc(map_entry_count,
+					      sizeof(struct hashmap_entry));
+
+	hashmap_init(&result->map, NULL, NULL, map_entry_count);
+	result->entries = entry;
+	for (p = line_begin; p + 1 < line_end; ++p, ++entry) {
+		c0 = *p;
+		c1 = *(p + 1);
+		/* Ignore whitespace pairs */
+		if (isspace(c0) && isspace(c1))
+			continue;
+		hash = tolower(c0) | (tolower(c1) << 8);
+		hashmap_entry_init(entry, hash);
+		hashmap_put(&result->map, entry);
+	}
+}
+
+static void free_fingerprint(struct fingerprint *f)
+{
+	hashmap_free(&f->map, 0);
+	free(f->entries);
+}
+
+static int fingerprint_similarity(struct fingerprint *a,
+				  struct fingerprint *b)
+{
+	int intersection = 0;
+	struct hashmap_iter iter;
+	struct hashmap_entry *entry;
+
+	hashmap_iter_init(&b->map, &iter);
+
+	while ((entry = hashmap_iter_next(&iter))) {
+		if (hashmap_get(&a->map, entry, NULL))
+			++intersection;
+	}
+	return intersection;
+}
+
+static void get_line_fingerprints(struct fingerprint *fingerprints,
+				  const char *content,
+				  const int *line_starts,
+				  int first_line,
+				  int nr_lines)
+{
+	int i;
+
+	line_starts += first_line;
+	for (i = 0; i < nr_lines; ++i) {
+		const char *linestart = content + line_starts[i];
+		const char *lineend = content + line_starts[i + 1];
+
+		get_fingerprint(fingerprints + i, linestart, lineend);
+	}
+}
+
+static void free_chunk_fingerprints(struct fingerprint *fingerprints,
+				    int nr_fingerprints)
+{
+	int i;
+
+	for (i = 0; i < nr_fingerprints; i++)
+		free_fingerprint(&fingerprints[i]);
+}
+
 static void fill_origin_fingerprints(struct blame_origin *o, mmfile_t *file)
 {
 	int *line_starts;
@@ -348,14 +426,16 @@ static void fill_origin_fingerprints(struct blame_origin *o, mmfile_t *file)
 		return;
 	o->num_lines = find_line_starts(&line_starts, o->file.ptr,
 					o->file.size);
-	/* TODO: Will fill in fingerprints in a future commit */
 	o->fingerprints = xcalloc(sizeof(struct fingerprint), o->num_lines);
+	get_line_fingerprints(o->fingerprints, o->file.ptr, line_starts,
+			      0, o->num_lines);
 	free(line_starts);
 }
 
 static void drop_origin_fingerprints(struct blame_origin *o)
 {
 	if (o->fingerprints) {
+		free_chunk_fingerprints(o->fingerprints, o->num_lines);
 		o->num_lines = 0;
 		FREE_AND_NULL(o->fingerprints);
 	}
@@ -935,27 +1015,69 @@ static int are_lines_adjacent(struct blame_line_tracker *first,
 	       first->s_lno + 1 == second->s_lno;
 }
 
+static void scan_parent_range(struct fingerprint *p_fps,
+			      struct fingerprint *t_fps, int t_idx,
+			      int from, int nr_lines,
+			      int *best_sim_val, int *best_sim_idx)
+{
+	int sim, p_idx;
+
+	for (p_idx = from; p_idx < from + nr_lines; p_idx++) {
+		sim = fingerprint_similarity(&t_fps[t_idx], &p_fps[p_idx]);
+		if (sim < *best_sim_val)
+			continue;
+		/* Break ties with the closest-to-target line number */
+		if (sim == *best_sim_val && *best_sim_idx != -1 &&
+		    abs(*best_sim_idx - t_idx) < abs(p_idx - t_idx))
+			continue;
+		*best_sim_val = sim;
+		*best_sim_idx = p_idx;
+	}
+}
+
 /*
- * This cheap heuristic assigns lines in the chunk to their relative location in
- * the parent's chunk.  Any additional lines are left with the target.
+ * The CHUNK threshold is for how similar we must be within a diff chunk, which
+ * is typically the adjacent '-' and '+' sections in a diff, separated by the
+ * ' ' context.
+ *
+ * We have a greater threshold for similarity for lines in any part of the
+ * parent's file.  If no line in the parent meets the appropriate threshold,
+ * then the blame_entry will stay with the target and be considered
+ * 'unblamable'.
  */
+#define FINGERPRINT_CHUNK_THRESHOLD	1
+#define FINGERPRINT_FILE_THRESHOLD	10
+
 static void guess_line_blames(struct blame_entry *e,
 			      struct blame_origin *parent,
 			      struct blame_origin *target,
 			      int offset, int parent_slno, int parent_len,
 			      struct blame_line_tracker *line_blames)
 {
-	int i, parent_idx;
+	int i, target_idx;
 
 	for (i = 0; i < e->num_lines; i++) {
-		parent_idx = e->s_lno + i + offset;
-		if (parent_slno <= parent_idx &&
-		    parent_idx < parent_slno + parent_len) {
+		int best_val = FINGERPRINT_CHUNK_THRESHOLD;
+		int best_idx = -1;
+
+		target_idx = e->s_lno + i;
+		scan_parent_range(parent->fingerprints,
+				  target->fingerprints, target_idx,
+				  parent_slno, parent_len,
+				  &best_val, &best_idx);
+		if (best_idx == -1) {
+			best_val = FINGERPRINT_FILE_THRESHOLD;
+			scan_parent_range(parent->fingerprints,
+					  target->fingerprints, target_idx,
+					  0, parent->num_lines,
+					  &best_val, &best_idx);
+		}
+		if (best_idx >= 0) {
 			line_blames[i].is_parent = 1;
-			line_blames[i].s_lno = parent_idx;
+			line_blames[i].s_lno = best_idx;
 		} else {
 			line_blames[i].is_parent = 0;
-			line_blames[i].s_lno = e->s_lno + i;
+			line_blames[i].s_lno = target_idx;
 		}
 	}
 }
Signature · 1 line
2.21.0.392.gf8f6787159e-goog
#6 lore raw

Re: [PATCH v6 3/6] blame: add the ability to ignore commits and their changes

In reply to Barret Rhoden

On Wed, Apr 10 2019, Barret Rhoden wrote:

(Just skimming)

Quoted 2 lines

revisions for commits that perform mass reformatting, and their users have the optional to ignore all of the commits in that file.

s/have the optional/have the option/

Review

Quoted 3 lines

+--ignore-revs-file <file>::

(quoted diff) +2 -0
+	Ignore revisions listed in `file`, one unabbreviated object name per line.
+	Whitespace and comments beginning with `#` are ignored.

Maybe just say "Ignore revisions listed in file, which is expected to be in the same format as an fsck.skipList.".

Documentation/blame-options.txt commit. And the default value is 40. If there are more than one · line 125

Quoted 2 lines
Documentation/blame-options.txt +2 -0
+	the `blame.ignoreRevsFile` config option.  An empty file name, `""`, will
+	clear the list of revs from previously processed files.

Maybe I haven't read this carefully enough but the use-case for this doesn't seem to be explained, you need this for the option, but the config file too? If I want to override fsck.skipList I do fsck.skipList=/dev/zero. Isn't that enough for this use-case without introducing config state-machine magic?

blame.c static void split_overlap(struct blame_entry *split, · line 738

Quoted 3 lines
blame.c +3 -0
+	split[0].unblamable = e->unblamable;
+	split[1].unblamable = e->unblamable;
+	split[2].unblamable = e->unblamable;

I wonder what the comfort level for people in general is before turning this sort of thing into a for-loop, 4? :)

blame.c static struct blame_entry *split_blame_at(struct blame_entry *e, int len, · line 935

Quoted 1 line
blame.c +1 -0
+	nr_lines = e->num_lines;	// e changes in the loop

A C++-like trailing comment.

t/t8013-blame-ignore-revs.sh line 33

Quoted 7 lines
t/t8013-blame-ignore-revs.sh new file +7 -0
+	grep "^[0-9a-f]\+ [0-9]\+ 1" blame_raw | sed -e "s/ .*//" >actual &&
+	git rev-parse X >expect &&
+	test_cmp expect actual &&
+
+	grep "^[0-9a-f]\+ [0-9]\+ 2" blame_raw | sed -e "s/ .*//" >actual &&
+	git rev-parse X >expect &&
+	test_cmp expect actual

The grep here is a bug. See my 4abf20f004 ("tests: fix unportable "\?" and "\+" regex syntax", 2019-02-21).

#2 lore raw

Re: [PATCH v6 1/6] Move init_skiplist() outside of fsck

In reply to Barret Rhoden
Quoted 3 lines · Barret Rhoden

init_skiplist() took a file consisting of SHA-1s and comments and added the objects to an oidset. This functionality is useful for other commands.

This change would be much easier to review if you led with a commit where you s/Invalid SHA-1/invalid object name/ (lower-case while we're at it), s/skip list/object name/ etc, and did that rename of the "hash" to "name" variable if you're so inclined.

Then you'd end up with a small refactoring change that changes the tests (or even just make the tests grep for e.g. "Could not open.*:
does-not-exist" instead), and the moving of the function would be entirely caught by the rename detection.

#11 lore raw

Re: [PATCH v6 4/6] blame: add config options to handle output for ignored lines

In reply to Barret Rhoden
Quoted 8 lines · Barret Rhoden

Sometimes we are unable to even guess at what commit touched a line. These lines are 'unblamable.' The second option, blame.maskIgnoredUnblamables, will zero the hash of any unblamable line.

For example, say we ignore e5e8d36d04cbe:

	e5e8d36d04cbe (Barret Rhoden  2016-04-11 13:57:54 -0400 26)
appears as:
	0000000000000 (Barret Rhoden  2016-04-11 13:57:54 -0400 26)

Wouldn't this make it impossible to tell between what's done by such a commit that was marked to be ignored, and what's done locally only in the working tree, which the users have long accustomed to see with the ^0*$ object name? I think it would make a lot more sense to show the object name of the "ignored" commit, which would be recognizable by the user who fed such an object name to the command in the first place. Alternatively, perhaps the same idea as replacing one of the hexdigits with '*' used by the other configuration can be applied to this as well?

#18 lore raw

Re: [PATCH v6 6/6] blame: use a fingerprint heuristic to match ignored lines

In reply to Barret Rhoden
Quoted 16 lines · Barret Rhoden

This replaces the heuristic used to identify lines from ignored commits with one that finds likely candidate lines in the parent's version of the file.

The old heuristic simply assigned lines in the target to the same line number (plus offset) in the parent. The new function uses a fingerprinting algorithm to detect similarity between lines.

The fingerprint code and the idea to use them for blame came from Michael Platings <[email protected]>.

For each line changed in the target, i.e. in a blame_entry touched by a target's diff, guess_line_blames() finds the best line in the parent, above a magic threshold. Ties are broken by proximity of the parent line number to the target's line.

Quoted 5 lines

We actually make two passes. The first pass checks in the diff chunk associated with the blame entry - specifically from blame_chunk(). Often times, those diff chunks are small; any 'context' in a normal diff chunk is broken up into multiple calls to blame_chunk(). We make a second pass over the entire parent, with a slightly higher threshold.

Two thoughts.

- Unless the 'old heuristic' is still available as an option after this step, a series that first begins with the 'old heuristic' and then later replaces it with the 'new heuristic' feels somewhat wasteful of reviewer resources, as the 'old heuristic' does not contribute an iota to the end result.

It is OK while the series is still in RFC/WIP stage, though. But because I got an impression that this is close to completion, so...

- I wonder if the hash used here can replace what is used in diffcore-delta.c as an improvement (or obviously vice versa), as using two (or more) ad-hoc fingerprinting function without having a clear reason why we need two instead of a unified one feels like a bad idea.

#19 lore raw

Re: [PATCH v6 6/6] blame: use a fingerprint heuristic to match ignored lines

Quoted 5 lines

- I wonder if the hash used here can replace what is used in diffcore-delta.c as an improvement (or obviously vice versa), as using two (or more) ad-hoc fingerprinting function without having a clear reason why we need two instead of a unified one feels like a bad idea.

Hi Junio,
If I understand correctly, the algorithm in diffcore-delta.c is intended to match files that contain identical lines (or 64-byte chunks). The fingerprinting that Barret & I are talking about is intended to match lines that contain identical byte pairs. With significant refactoring, you could make the diffcore-delta algorithm apply in both cases but I think the end result would be longer and more complicated than keeping the two separate. Unlike hashing a line, hashing a byte pair is trivial. Unlike hashing lines, all except the first and last bytes are included in two "hashes" - "hello" is hashed to "he", "el", "ll", "lo".
So based on my limited understanding of diffcore-delta.c I think the two are algorithms are sufficiently different in intent and in implementation that it's appropriate to keep them separate.

Regarding the "old heuristic" I think there may still be a use case for that but I'll expand on that later.

Thanks,
-Michael

#12 lore raw

Re: [PATCH v6 4/6] blame: add config options to handle output for ignored lines

In reply to Junio C Hamano
Quoted 9 lines · Junio C Hamano

Wouldn't this make it impossible to tell between what's done by such a commit that was marked to be ignored, and what's done locally only in the working tree, which the users have long accustomed to see with the ^0*$ object name? I think it would make a lot more sense to show the object name of the "ignored" commit, which would be recognizable by the user who fed such an object name to the command in the first place. Alternatively, perhaps the same idea as replacing one of the hexdigits with '*' used by the other configuration can be applied to this as well?

I had the same objection to zeroing out hashes, but this option is off by default so I think it's OK.
If you enable both blame.markIgnoredLines and blame.maskIgnoredUnblamables then the hash does appear as "*0000000000" like you suggest. I think it's appropriate that the '*' is only added if you opt in with the markIgnoredLines option.

If you only enable blame.markIgnoredLines then the hash for "unblamable" lines appears as e.g. "*3252488f5" - this doesn't seem right to me because the commit *wasn't* ignored, it is in fact the commit in which that line was added. I think '*' should denote "this information may be inaccurate" as that's what a typical user needs to be aware of. However given that "unblamable" lines tend to be either empty or a single character I'm not going to insist :)

#13 lore raw

Re: [PATCH v6 4/6] blame: add config options to handle output for ignored lines

Quoted 3 lines · Michael Platings

If you only enable blame.markIgnoredLines then the hash for "unblamable" lines appears as e.g. "*3252488f5" - this doesn't seem right to me because the commit *wasn't* ignored,

I think you misunderstood me. I was merely suggesting to use the approach to mark the line in a way other than using the NULLed out object name that has been reserved for something totally different, and hinting with "the same *idea*".

And that idea is not even original to this series; the "^" marker that is used to say "the line is attributed to this commit, but that may only be because you blamed with commit range A..B and we reached the bottom of the range---if you dug further, you might find the line originates from another commit" is the origin of the same idea, and this topic borrows it and uses a different mark, i.e. '*', for the "we are not certain---take this with grain of salt" mark.

If you ended up hitting the commit the user wanted to ignore, perhaps you can find another character that is different from '^' or '*' and use that, following the same idea.

That is what I meant. So you shouldn't be worried about using the same '*' making the result ambiguous.

By the way, a configuration only feature is something we usually do not accept. A feature must be guarded with --command-line-option and then optionally can have a corresponding configuration once the option proves to be useful enough that it becomes useful to be able to say "in this repository (or to this user), the feature is on by default".

#7 lore raw

Re: [PATCH v6 3/6] blame: add the ability to ignore commits and their changes

In reply to Ævar Arnfjörð Bjarmason
Quoted 8 lines

Documentation/blame-options.txt commit. And the default value is 40. If there are more than one · line 125

Quoted 2 lines
Documentation/blame-options.txt +2 -0
+     the `blame.ignoreRevsFile` config option.  An empty file name, `""`, will
+     clear the list of revs from previously processed files.

Maybe I haven't read this carefully enough but the use-case for this doesn't seem to be explained, you need this for the option, but the config file too? If I want to override fsck.skipList I do fsck.skipList=/dev/zero. Isn't that enough for this use-case without introducing config state-machine magic?

The difference between blame.ignoreRevsFile and fsck.skipList is that ignoreRevsFile can be specified repeatedly. This is useful if you have one file listing reformatting commits, another listing renaming commits etc. Or maybe a checked-in list of commits to ignore, and a personal list of commits to ignore. However sometimes you're going to want to *not* ignore those commits, so you need a way to discard the previously specified options. To accommodate all operating systems an empty string seems the best way to do this.

#14 lore raw

Re: [PATCH v6 4/6] blame: add config options to handle output for ignored lines

In reply to Junio C Hamano
Quoted 8 lines · Junio C Hamano
Quoted 3 lines

If you only enable blame.markIgnoredLines then the hash for "unblamable" lines appears as e.g. "*3252488f5" - this doesn't seem right to me because the commit *wasn't* ignored,

I think you misunderstood me. I was merely suggesting to use the approach to mark the line in a way other than using the NULLed out object name that has been reserved for something totally different, and hinting with "the same *idea*".

Hi Junio, that paragraph wasn't targetted at yourself, more a comment on the functionality as it exists in the latest patch series. Sorry for not making that clear.

Quoted 7 lines

the "^" marker
that is used to say "the line is attributed to this commit, but that may only be because you blamed with commit range A..B and we reached the bottom of the range---if you dug further, you might find the line originates from another commit" is the origin of the same idea, and this topic borrows it and uses a different mark, i.e. '*', for the "we are not certain---take this with grain of salt" mark.

So it sounds like we have many types of blame to consider:

1) This commit is truly the last one to touch this line, and you didn't ask to ignore it.
2) This commit is truly the last one to touch this line, but you asked to ignore it (AKA "unblamable").
3) This commit is at the bottom of the range of commits (^)
4) The "true" commit was ignored but we guess this is the one you're actually interested in (*)
5) The "true" commit was ignored and we've reached the bottom of the range of commits (^*)?
6) This commit is at the bottom of the range of commits, and you asked to ignore it.

Quoted 3 lines

If you ended up hitting the commit the user wanted to ignore, perhaps you can find another character that is different from '^' or '*' and use that, following the same idea.

I personally don't find the "unblamable" lines interesting enough to justify giving them a symbol. But if Barret strongly feels that such lines should get a '*' then I won't fight it - these lines tend to be as simple as "}".

Quoted 6 lines

By the way, a configuration only feature is something we usually do not accept. A feature must be guarded with --command-line-option and then optionally can have a corresponding configuration once the option proves to be useful enough that it becomes useful to be able to say "in this repository (or to this user), the feature is on by default".

In that case we definitely need a --mark-ignored-lines option to git blame, and I would strongly prefer that we also keep the blame.markIgnoredLines option as I for one will be switching it on.

#22 lore raw
In reply to Barret Rhoden

Hi Barret,

This works pretty well for the typical reformatting use case now. I've run it over every commit of every .c file in the git project root, both forwards and backwards with every combination of -w/-M/-C and can't get it to crash so I think it's good in that respect.

However, it can still attribute lines to the wrong parent line. See
https://pypi.org/project/autopep8/#usage for an example reformatting that it gets a bit confused on. The patch I submitted handles this case correctly because it uses information about the more similar lines to decide how more ambiguous lines should be matched.

You also gave an example of:

        commit-a 11) void new_func_1(void *x, void *y);
        commit-b 12) void new_func_2(void *x, void *y);

Being reformatted to:

        commit-a 11) void new_func_1(void *x,
        commit-b 12)                 void *y);
        commit-b 13) void new_func_2(void *x,
        commit-b 14)                 void *y);

The patch I submitted handles this case correctly, assigning line 12 to commit-a because it scales the parent line numbers according to the relative diff chunk sizes instead of assuming a 1-1 mapping.

So I do ask that you incorporate more of my patch, including the test code. It is more complex but I hope this demonstrates that there are reasons for that. Happy to provide more examples or explanation if it would help. On the other hand if you have examples where it falls short then I'd be interested to know.

The other major use case that I'm interested in is renaming. In this case, the git-hyper-blame approach of mapping line numbers 1-1 works perfectly. Here's an example. Before:

        commit-a 11) Position MyClass::location(Offset O) {
        commit-b 12)    return P + O;
        commit-c 13) }

After:

        commit-a 11) Position MyClass::location(Offset offset) {
        commit-a 12)    return position + offset;
        commit-c 13) }

With the fuzzy matching, line 12 gets incorrectly matched to parent line 11 because the similarity of "position" and "offset" outweighs the similarity of "return". I'm considering adding even more complexity to my patch such that parts of a line that have already been matched can't be matched again by other lines.

But the other possibility is that we let the user choose the heuristic. For a commit where they know that line numbers haven't changed they could choose 1-1 matching, while for a reformatting commit they could use fuzzy matching. I welcome your thoughts.

-Michael

#23 lore raw

Hi Michael -

Quoted 7 lines · Michael Platings

Hi Barret,

This works pretty well for the typical reformatting use case now. I've run it over every commit of every .c file in the git project root, both forwards and backwards with every combination of -w/-M/-C and can't get it to crash so I think it's good in that respect.

Quoted 5 lines

However, it can still attribute lines to the wrong parent line. See
https://pypi.org/project/autopep8/#usage for an example reformatting that it gets a bit confused on. The patch I submitted handles this case correctly because it uses information about the more similar lines to decide how more ambiguous lines should be matched.

Yeah - I ran your tests against it and noticed a few cases weren't handled.

Quoted 16 lines

You also gave an example of:

         commit-a 11) void new_func_1(void *x, void *y);
         commit-b 12) void new_func_2(void *x, void *y);

Being reformatted to:

         commit-a 11) void new_func_1(void *x,
         commit-b 12)                 void *y);
         commit-b 13) void new_func_2(void *x,
         commit-b 14)                 void *y);

The patch I submitted handles this case correctly, assigning line 12 to commit-a because it scales the parent line numbers according to the relative diff chunk sizes instead of assuming a 1-1 mapping.

Quoted 5 lines

So I do ask that you incorporate more of my patch, including the test code. It is more complex but I hope this demonstrates that there are reasons for that. Happy to provide more examples or explanation if it would help. On the other hand if you have examples where it falls short then I'd be interested to know.

My main concerns:
- Can your version reach outside of a diff chunk? such as in my "header moved" case. That was a simplified version of something that pops up in a major file reformatting of mine, where a "return 0;" was matched as context and broke a diff chunk up into two blame_chunk() calls. I tend to think of this as the "split diff chunk."

- Complexity and possibly performance. The recursive stuff made me wonder about it a bit. It's no reason not to use it, just need to check it more closely.

Is the latest version of your stuff still the one you posted last week or so? If we had a patch applied onto this one with something like an ifdef or a dirt-simple toggle, we can play with both of them in the same codebase.

Similarly, do you think the "two pass" approach I have (check the chunk, then check the parent file) would work with your recursive partitioning style? That might make yours able to handle the "split diff chunk" case.

Quoted 20 lines

The other major use case that I'm interested in is renaming. In this case, the git-hyper-blame approach of mapping line numbers 1-1 works perfectly. Here's an example. Before:

         commit-a 11) Position MyClass::location(Offset O) {
         commit-b 12)    return P + O;
         commit-c 13) }

After:

         commit-a 11) Position MyClass::location(Offset offset) {
         commit-a 12)    return position + offset;
         commit-c 13) }

With the fuzzy matching, line 12 gets incorrectly matched to parent line 11 because the similarity of "position" and "offset" outweighs the similarity of "return". I'm considering adding even more complexity to my patch such that parts of a line that have already been matched can't be matched again by other lines.

Quoted 4 lines

But the other possibility is that we let the user choose the heuristic. For a commit where they know that line numbers haven't changed they could choose 1-1 matching, while for a reformatting commit they could use fuzzy matching. I welcome your thoughts.

No algorithm will work for all cases. The one you just gave had the simple heuristic working better than a complex one. We could make it more complex, but then another example may be worse. I can live with some inaccuracy in exchange for simplicity.

I ran into something similar with the THRESHOLD #defines. You want it to be able to match certain things, but not other things. How similar does something have to be? Should it depend on how far away the matching line is from the source line? I went with a "close enough is good enough" approach, since we're marking with a '*' or something anyways, so the user should know to not trust it 100%.

Thanks,

Barret

#3 lore raw

Re: [PATCH v6 1/6] Move init_skiplist() outside of fsck

In reply to Ævar Arnfjörð Bjarmason
Quoted 12 lines · Ævar Arnfjörð Bjarmason
Quoted 3 lines · Barret Rhoden

init_skiplist() took a file consisting of SHA-1s and comments and added the objects to an oidset. This functionality is useful for other commands.

This change would be much easier to review if you led with a commit where you s/Invalid SHA-1/invalid object name/ (lower-case while we're at it), s/skip list/object name/ etc, and did that rename of the "hash" to "name" variable if you're so inclined.

Quoted 5 lines

Then you'd end up with a small refactoring change that changes the tests (or even just make the tests grep for e.g. "Could not open.*:
does-not-exist" instead), and the moving of the function would be entirely caught by the rename detection.

Can do. I'll split this up in the next round.

Thanks,

Barret

#8 lore raw

Re: [PATCH v6 3/6] blame: add the ability to ignore commits and their changes

In reply to Michael Platings
Quoted 9 lines · Michael Platings
Quoted 8 lines

Documentation/blame-options.txt commit. And the default value is 40. If there are more than one · line 125

Quoted 2 lines
Documentation/blame-options.txt +2 -0
+     the `blame.ignoreRevsFile` config option.  An empty file name, `""`, will
+     clear the list of revs from previously processed files.

Maybe I haven't read this carefully enough but the use-case for this doesn't seem to be explained, you need this for the option, but the config file too? If I want to override fsck.skipList I do fsck.skipList=/dev/zero. Isn't that enough for this use-case without introducing config state-machine magic?

Quoted 9 lines

The difference between blame.ignoreRevsFile and fsck.skipList is that ignoreRevsFile can be specified repeatedly. This is useful if you have one file listing reformatting commits, another listing renaming commits etc. Or maybe a checked-in list of commits to ignore, and a personal list of commits to ignore. However sometimes you're going to want to *not* ignore those commits, so you need a way to discard the previously specified options. To accommodate all operating systems an empty string seems the best way to do this.

In a previous round of reviews[1], this style was recommended. It's based on what credential.helper does.

The main thing I've been using the --ignore-revs-file="" for is to turn off my default ignore list for debugging. =)

Thanks,

Barret

[1] https://public-inbox.org/git/[email protected]/

#9 lore raw

Re: [PATCH v6 3/6] blame: add the ability to ignore commits and their changes

In reply to Ævar Arnfjörð Bjarmason

On 4/10/19 3:00 PM, Ævar Arnfjörð Bjarmason wrote:
[snip]

Quoted 6 lines

blame.c static void split_overlap(struct blame_entry *split, · line 738

Quoted 3 lines
blame.c +3 -0
+	split[0].unblamable = e->unblamable;
+	split[1].unblamable = e->unblamable;
+	split[2].unblamable = e->unblamable;

I wonder what the comfort level for people in general is before turning this sort of thing into a for-loop, 4? :)

4 sounds good to me. =)

Quoted 4 lines

blame.c static struct blame_entry *split_blame_at(struct blame_entry *e, int len, · line 935

Quoted 1 line
blame.c +1 -0
+	nr_lines = e->num_lines;	// e changes in the loop

A C++-like trailing comment.

Quoted 10 lines

t/t8013-blame-ignore-revs.sh line 33

Quoted 7 lines
t/t8013-blame-ignore-revs.sh new file +7 -0
+	grep "^[0-9a-f]\+ [0-9]\+ 1" blame_raw | sed -e "s/ .*//" >actual &&
+	git rev-parse X >expect &&
+	test_cmp expect actual &&
+
+	grep "^[0-9a-f]\+ [0-9]\+ 2" blame_raw | sed -e "s/ .*//" >actual &&
+	git rev-parse X >expect &&
+	test_cmp expect actual

The grep here is a bug. See my 4abf20f004 ("tests: fix unportable "\?" and "\+" regex syntax", 2019-02-21).

Thanks - will fix up this stuff in the next round.

#15 lore raw

Re: [PATCH v6 4/6] blame: add config options to handle output for ignored lines

In reply to Michael Platings

Hi -

Quoted 36 lines · Michael Platings
Quoted 8 lines · Junio C Hamano
Quoted 3 lines

If you only enable blame.markIgnoredLines then the hash for "unblamable" lines appears as e.g. "*3252488f5" - this doesn't seem right to me because the commit *wasn't* ignored,

I think you misunderstood me. I was merely suggesting to use the approach to mark the line in a way other than using the NULLed out object name that has been reserved for something totally different, and hinting with "the same *idea*".

Hi Junio, that paragraph wasn't targetted at yourself, more a comment on the functionality as it exists in the latest patch series. Sorry for not making that clear.

Quoted 7 lines

the "^" marker
that is used to say "the line is attributed to this commit, but that may only be because you blamed with commit range A..B and we reached the bottom of the range---if you dug further, you might find the line originates from another commit" is the origin of the same idea, and this topic borrows it and uses a different mark, i.e. '*', for the "we are not certain---take this with grain of salt" mark.

So it sounds like we have many types of blame to consider:

1) This commit is truly the last one to touch this line, and you didn't ask to ignore it.
2) This commit is truly the last one to touch this line, but you asked to ignore it (AKA "unblamable").
3) This commit is at the bottom of the range of commits (^)
4) The "true" commit was ignored but we guess this is the one you're actually interested in (*)
5) The "true" commit was ignored and we've reached the bottom of the range of commits (^*)?
6) This commit is at the bottom of the range of commits, and you asked to ignore it.

Quoted 8 lines
Quoted 3 lines

If you ended up hitting the commit the user wanted to ignore, perhaps you can find another character that is different from '^' or '*' and use that, following the same idea.

I personally don't find the "unblamable" lines interesting enough to justify giving them a symbol. But if Barret strongly feels that such lines should get a '*' then I won't fight it - these lines tend to be as simple as "}".

I'm fine with not zeroing the hash, so long as there's some way to mark it.

We could mark with another *, such that if we mark-ignored and mark-unblamable you get "**hash". You can't have an unblamable that isn't from an ignored commit, so a single '*' has only one meaning, based on your config options.

If that works for you all, I can change this to markIgnoredUnblamables (instead of 'mask') in the next version.

Quoted 10 lines
Quoted 6 lines

By the way, a configuration only feature is something we usually do not accept. A feature must be guarded with --command-line-option and then optionally can have a corresponding configuration once the option proves to be useful enough that it becomes useful to be able to say "in this repository (or to this user), the feature is on by default".

In that case we definitely need a --mark-ignored-lines option to git blame, and I would strongly prefer that we also keep the blame.markIgnoredLines option as I for one will be switching it on.

I'd also keep this set. I think the whole reason for these config options was that everyone has a different preference, but that preference rarely changes. I don't want to have to type --mark-ignored-lines every time I run git blame. If I had to, I'd have to alias git blame or something.

I think having config options for these sorts of things is fine, since we know already that for a given user+repo, we want the feature on (or off). But if I have to remove it, then let me know.

Thanks,

Barret

#20 lore raw

Re: [PATCH v6 6/6] blame: use a fingerprint heuristic to match ignored lines

In reply to Junio C Hamano
Quoted 10 lines · Junio C Hamano

Two thoughts.

- Unless the 'old heuristic' is still available as an option after this step, a series that first begins with the 'old heuristic' and then later replaces it with the 'new heuristic' feels somewhat wasteful of reviewer resources, as the 'old heuristic' does not contribute an iota to the end result.

It is OK while the series is still in RFC/WIP stage, though. But because I got an impression that this is close to completion, so...

Can do. I wasn't sure yet where things were going, but in the final version, I can yank out the old heuristic from the patch set.

Though the old heuristic is pretty basic - really just a couple lines - and it may help to see it before looking at a more complicated version. Especially since it helps break the commit up into "infrastructure to ignore commits" and "brains to find the right commit to blame" while still being functional between the commits.

Thanks,

Barret

#24 lore raw
In reply to Barret Rhoden
Quoted 2 lines

My main concerns:
- Can your version reach outside of a diff chunk?

Currently no. It's optimised for reformatting and renaming, both of which preserve ordering. I could look into allowing disordered matches where the similarity is high, while still being biased towards ordered matches. If you can post more examples that would be helpful.

Quoted 3 lines

- Complexity and possibly performance. The recursive stuff made me wonder about it a bit. It's no reason not to use it, just need to check it more closely.

Complexity I can't deny, I can only mitigate it with documentation/comments. I optimised the code pretty heavily and tested on some contrived worst-case scenarios and the performance was still good so I'm not worried about that.

Quoted 2 lines

Is the latest version of your stuff still the one you posted last week or so?

Yes. But reaching outside the chunk might lead to a significantly different API in the next version...

Quoted 3 lines

Similarly, do you think the "two pass" approach I have (check the chunk, then check the parent file) would work with your recursive partitioning style? That might make yours able to handle the "split diff chunk" case.

Yes, should do. I'll see what I can come up with this week.

Quoted 4 lines

No algorithm will work for all cases. The one you just gave had the simple heuristic working better than a complex one. We could make it more complex, but then another example may be worse. I can live with some inaccuracy in exchange for simplicity.

Exactly, no algorithm will work for all cases. So what I'm suggesting is that it might be best to let the user choose which heuristic is appropriate for a given commit. If they know that the simple heuristic works best then perhaps we should let them choose that rather than only offering a one-size-fits-all option. But if we do want to go for one-size-fits-all then I'm very keen to make sure it at least solves the specific cases that we know about.

#21 lore raw

Re: [PATCH v6 6/6] blame: use a fingerprint heuristic to match ignored lines

In reply to Barret Rhoden
Quoted 4 lines · Barret Rhoden

Though the old heuristic is pretty basic - really just a couple lines -
and it may help to see it before looking at a more complicated version.

OK, then.

Thanks.