| Message ID | 20200203210445.2854-2-johan@herland.net (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [1/2] t3305: check notes fanout more carefully and robustly | expand |
On 2020-02-03 at 21:04:45, Johan Herland wrote: > As noted in the previous commit, the nature of the fanout heuristic > in the notes code causes the exact point at which we increase or > decrease the notes fanout to vary with the objects being annotated. > Since the object ids generated by the test environment are > deterministic (by design), the notes generated and tested by t3305 > are always the same, and we therefore happen to see the same fanout > behavior from one run to the next. > > Coincidentally, if we were to change the test environment slightly > (say by making a test commit on an unrelated branch before we start > the t3305 test proper), we not only see the fanout switch happen at > different points, we also manage to trigger a _bug_ in the notes > code where the fanout 1 -> 0 switch is not applied uniformly across > the notes tree, but instead yields a notes tree like this: > > ... > bdeafb301e44b0e4db0f738a2d2a7beefdb70b70 > bff2d39b4f7122bd4c5caee3de353a774d1e632a > d3/8ec8f851adf470131178085bfbaab4b12ad2a7 > e0b173960431a3e692ae929736df3c9b73a11d5b > eb3c3aede523d729990ac25c62a93eb47c21e2e3 > ... > > The bug occurs when we are writing out a notes tree with a newly > decreased fanout, and the notes tree contains unexpanded subtrees > that should be consolidated into the parent tree as a consequence of > the decreased fanout): > > Subtrees that happen to sit at an _even_ level in the internal notes > 16-tree structure (in other words: subtrees whose path - "d3" in the > example above - is unique in the first nibble - i.e. there are no > other note paths that start with "d") are _not_ unpacked as part of > the tree writeout. This error will repeat itself in subsequent note > trees until the subtree is forced to be unpacked. In t3305 this only > happens when the d38ec8f8 note is itself removed from the tree. > > The error is not severe (no information is lost, and the notes code > is able to read/decode this tree and manipulate it correctly), but > this is nonetheless a bug in the current implementation that should > be fixed. > > That said, fixing the off-by-one error is not without complications: > We must take into account that the load_subtree() call from > for_each_note_helper() (that is now done to correctly unpack the > subtree while we're writing out the notes tree) may end up inserting > unpacked non-notes into the linked list of non_note entries held by > the struct notes_tree. Since we are in the process of writing out the > notes tree, this linked list is currently in the process of being > traversed by write_each_non_note_until(). The unpacked non-notes are > necessarily inserted between the last non-note we wrote out, and the > next non-note to be written. Hence, we cannot simply hold the > next_non_note to write in struct write_each_note_data (as we would > then silently skip these newly inserted notes), but must instead > always follow the ->next pointer from the last non-note we wrote. > (This part is was caught by an existing test in t3304.) I think you have "is was" here. You probably want one or the other. > Cc: Johannes Schindelin <Johannes.Schindelin@gmx.de> > Cc: Brian M. Carlson <sandals@crustytoothpaste.net> I generally write my name in lower case, but I think typically we prefer to omit Cc lines in patches (unlike LKML), so it may just be better to remove these lines. > Signed-off-by: Johan Herland <johan@herland.net> Patch 1 looked good to me. Your explanation here makes sense, but I must admit that I still don't understand this code, so I can't give an outright approval. I do appreciate that it comes with a test, though. I haven't tested, but I expect this series will make Dscho's patch unnecessary, so I'll drop it in my reroll unless one of you tells me to keep it.
On Tue, Feb 4, 2020 at 2:05 AM brian m. carlson <sandals@crustytoothpaste.net> wrote: > On 2020-02-03 at 21:04:45, Johan Herland wrote: > > always follow the ->next pointer from the last non-note we wrote. > > (This part is was caught by an existing test in t3304.) > > I think you have "is was" here. You probably want one or the other. Will fix. > > Cc: Johannes Schindelin <Johannes.Schindelin@gmx.de> > > Cc: Brian M. Carlson <sandals@crustytoothpaste.net> > > I generally write my name in lower case, but I think typically we prefer > to omit Cc lines in patches (unlike LKML), so it may just be better to > remove these lines. Ack. Will remove. > > Signed-off-by: Johan Herland <johan@herland.net> > > Patch 1 looked good to me. Your explanation here makes sense, but I > must admit that I still don't understand this code, so I can't give an > outright approval. I do appreciate that it comes with a test, though. Thanks for having a look. I must admit it's hard to get back into this code even though I originally wrote it. I've been trying to prepare a patch #3 which decreases the fanout more aggressively on notes removal (having a fanout of 1 in a tree of 17 notes is bit ridiculous, IMHO), but I'm not yet able to figure something out that behaves in a stable manner. (I find scenarios where removing a note will switch fanout from 1 to 0, but removing another note will then switch fanout from 0 back to 1, and so on, and the current notes code does not have these problems, AFAICS.) > I haven't tested, but I expect this series will make Dscho's patch > unnecessary, so I'll drop it in my reroll unless one of you tells me to > keep it. Yes, my patch includes Dscho's change. I don't particularly care whichever lands first, and I can easily rebase on top of yours. ...Johan
diff --git a/notes.c b/notes.c index 0c79964c26..2de7f4bcfb 100644 --- a/notes.c +++ b/notes.c @@ -576,16 +576,16 @@ static int for_each_note_helper(struct notes_tree *t, struct int_node *tree, * the note tree that have not yet been explored. There * is a direct relationship between subtree entries at * level 'n' in the tree, and the 'fanout' variable: - * Subtree entries at level 'n <= 2 * fanout' should be + * Subtree entries at level 'n < 2 * fanout' should be * preserved, since they correspond exactly to a fanout * directory in the on-disk structure. However, subtree - * entries at level 'n > 2 * fanout' should NOT be + * entries at level 'n >= 2 * fanout' should NOT be * preserved, but rather consolidated into the above * notes tree level. We achieve this by unconditionally * unpacking subtree entries that exist below the * threshold level at 'n = 2 * fanout'. */ - if (n <= 2 * fanout && + if (n < 2 * fanout && flags & FOR_EACH_NOTE_YIELD_SUBTREES) { /* invoke callback with subtree */ unsigned int path_len = @@ -602,7 +602,7 @@ static int for_each_note_helper(struct notes_tree *t, struct int_node *tree, path, cb_data); } - if (n > fanout * 2 || + if (n >= 2 * fanout || !(flags & FOR_EACH_NOTE_DONT_UNPACK_SUBTREES)) { /* unpack subtree and resume traversal */ tree->a[i] = NULL; @@ -723,13 +723,15 @@ static int write_each_note_helper(struct tree_write_stack *tws, struct write_each_note_data { struct tree_write_stack *root; - struct non_note *next_non_note; + struct non_note **nn_list; + struct non_note *nn_prev; }; static int write_each_non_note_until(const char *note_path, struct write_each_note_data *d) { - struct non_note *n = d->next_non_note; + struct non_note *p = d->nn_prev; + struct non_note *n = p ? p->next : *d->nn_list; int cmp = 0, ret; while (n && (!note_path || (cmp = strcmp(n->path, note_path)) <= 0)) { if (note_path && cmp == 0) @@ -740,9 +742,10 @@ static int write_each_non_note_until(const char *note_path, if (ret) return ret; } + p = n; n = n->next; } - d->next_non_note = n; + d->nn_prev = p; return 0; } @@ -1177,7 +1180,8 @@ int write_notes_tree(struct notes_tree *t, struct object_id *result) strbuf_init(&root.buf, 256 * (32 + the_hash_algo->hexsz)); /* assume 256 entries */ root.path[0] = root.path[1] = '\0'; cb_data.root = &root; - cb_data.next_non_note = t->first_non_note; + cb_data.nn_list = &(t->first_non_note); + cb_data.nn_prev = NULL; /* Write tree objects representing current notes tree */ flags = FOR_EACH_NOTE_DONT_UNPACK_SUBTREES | diff --git a/t/t3305-notes-fanout.sh b/t/t3305-notes-fanout.sh index 402057c83a..3b4753e1b4 100755 --- a/t/t3305-notes-fanout.sh +++ b/t/t3305-notes-fanout.sh @@ -30,6 +30,12 @@ all_notes_have_fanout() { done } +test_expect_success 'tweak test environment' ' + git checkout -b nondeterminism && + test_commit A && + git checkout --orphan with_notes; +' + test_expect_success 'creating many notes with git-notes' ' num_notes=300 && i=0 &&
As noted in the previous commit, the nature of the fanout heuristic in the notes code causes the exact point at which we increase or decrease the notes fanout to vary with the objects being annotated. Since the object ids generated by the test environment are deterministic (by design), the notes generated and tested by t3305 are always the same, and we therefore happen to see the same fanout behavior from one run to the next. Coincidentally, if we were to change the test environment slightly (say by making a test commit on an unrelated branch before we start the t3305 test proper), we not only see the fanout switch happen at different points, we also manage to trigger a _bug_ in the notes code where the fanout 1 -> 0 switch is not applied uniformly across the notes tree, but instead yields a notes tree like this: ... bdeafb301e44b0e4db0f738a2d2a7beefdb70b70 bff2d39b4f7122bd4c5caee3de353a774d1e632a d3/8ec8f851adf470131178085bfbaab4b12ad2a7 e0b173960431a3e692ae929736df3c9b73a11d5b eb3c3aede523d729990ac25c62a93eb47c21e2e3 ... The bug occurs when we are writing out a notes tree with a newly decreased fanout, and the notes tree contains unexpanded subtrees that should be consolidated into the parent tree as a consequence of the decreased fanout): Subtrees that happen to sit at an _even_ level in the internal notes 16-tree structure (in other words: subtrees whose path - "d3" in the example above - is unique in the first nibble - i.e. there are no other note paths that start with "d") are _not_ unpacked as part of the tree writeout. This error will repeat itself in subsequent note trees until the subtree is forced to be unpacked. In t3305 this only happens when the d38ec8f8 note is itself removed from the tree. The error is not severe (no information is lost, and the notes code is able to read/decode this tree and manipulate it correctly), but this is nonetheless a bug in the current implementation that should be fixed. That said, fixing the off-by-one error is not without complications: We must take into account that the load_subtree() call from for_each_note_helper() (that is now done to correctly unpack the subtree while we're writing out the notes tree) may end up inserting unpacked non-notes into the linked list of non_note entries held by the struct notes_tree. Since we are in the process of writing out the notes tree, this linked list is currently in the process of being traversed by write_each_non_note_until(). The unpacked non-notes are necessarily inserted between the last non-note we wrote out, and the next non-note to be written. Hence, we cannot simply hold the next_non_note to write in struct write_each_note_data (as we would then silently skip these newly inserted notes), but must instead always follow the ->next pointer from the last non-note we wrote. (This part is was caught by an existing test in t3304.) Cc: Johannes Schindelin <Johannes.Schindelin@gmx.de> Cc: Brian M. Carlson <sandals@crustytoothpaste.net> Signed-off-by: Johan Herland <johan@herland.net> --- notes.c | 20 ++++++++++++-------- t/t3305-notes-fanout.sh | 6 ++++++ 2 files changed, 18 insertions(+), 8 deletions(-)