BUG: git diff --name-only ignores --ignore-*

BUG: git diff --name-only ignores --ignore-*

[Stefan Hoffmeister]

When using the --name-only option on git diff (git 2.33.0),
command-line parameters such as  --ignore-blank-lines are not
respected.

Instead, git diff behaves as if these options were not specified at all.

This is either an implementation defect, or a documentation defect,
because that (IMHO surprising) behaviour is not spelled out in
https://git-scm.com/docs/git-diff.

IMHO, this is an implementation defect, as --ignore-* is orthogonal to
--name-only. I would like to use --name-only to identify the set of
files which have only(!) --ignore-* changes (by intersecting the set
of files returned by git diff with, and without --ignore-*).

The bash script below reproduces this for --ignore-blank-lines; the
same would extend to the other --ignore-* options as well.

****************************
#!/usr/bin/env bash

set -e

GIT_REPO=repo
TEST_FILE=the_file.txt

rm -rf ./${GIT_REPO}
mkdir ${GIT_REPO}
pushd ${GIT_REPO}

git init --initial-branch=main
git config user.name "myuser"
git config user.email "myuser@example.com"

cat << EOF > ${TEST_FILE}
Hello world.
EOF

git add -A
git commit -m"Initial commit"

# add empty lines
cat << EOF >> ${TEST_FILE}


EOF

without_name_only=$(git diff --ignore-blank-lines)
echo "----
Without --name-only:
${without_name_only}"

with_name_only=$(git diff --ignore-blank-lines --name-only)
echo "----
With --name-only:
${with_name_only}"

if [[ ${without_name_only} != ${with_name_only} ]] ; then
    echo "
fail: expected same data returned
"
    exit 1
else
    echo "
pass: all good
"
    exit 0
fi

Re: BUG: git diff --name-only ignores --ignore-*

[Junio C Hamano]

Stefan Hoffmeister <stefan.hoffmeister@gmail.com> writes:

> When using the --name-only option on git diff (git 2.33.0),
> command-line parameters such as  --ignore-blank-lines are not
> respected.

I think I've seen something related this one at around

  https://lore.kernel.org/git/xmqq4kkl1atq.fsf@gitster.c.googlers.com/

(look for mention of --name-only in the message).

It would make things more expensive to make --name-only inspect
changes when these options are in effect (iow, the --name-only
currently is about "are these the same blob object?  if not, show
the path" and nothing else) but it would probably be OK, as the user
expresses their willingness to spend extra cycles by explicitly
giving --ignore-* options.

Re: BUG: git diff --name-only ignores --ignore-*

[Stefan Hoffmeister]

A quick peek suggests options to tackle this:

a) by including XDF_IGNORE_BLANK_LINES as an option that also forces a
content check
b) adjusting the implementation in diff_flush to better take into
account the specified --ignore-* flags

Now, looking closely at diff_flush it would seem as if all of the
command-line options
  --raw
  --name-only
  --name-status
might benefit from taking into account --ignore-*

So, in the end this is a matter intent and requirements - what should
happen whenever a user provided --ignore*?

From my point of view, git diff --raw, --name-only, name-status all
should honour --ignore-whitespace*, and all should also honour
--ignore-blank-lines

Below please find a patch which shows the code areas affected;
implementation quality is not acceptable as-is (logic distributed
across the function), because I'd love to get some feedback on --raw
and --name-status and the general approach.

*********************

diff --git a/diff.c b/diff.c
index a8113f1707..7b19cfa5f0 100644
--- a/diff.c
+++ b/diff.c
@@ -4645,11 +4645,13 @@ void diff_setup_done(struct diff_options *options)
/*
* Most of the time we can say "there are changes"
* only by checking if there are changed paths, but
- * --ignore-whitespace* options force us to look
- * inside contents.
+ * options
+ * --ignore-whitespace*
+ * --ignore-blank-lines
+ * force us to look inside contents.
*/
- if ((options->xdl_opts & XDF_WHITESPACE_FLAGS) ||
+ if ((options->xdl_opts & (XDF_WHITESPACE_FLAGS | XDF_IGNORE_BLANK_LINES)) ||
options->ignore_regex_nr)
options->flags.diff_from_contents = 1;
else
@@ -6408,10 +6410,10 @@ void diff_flush(struct diff_options *options)
if (!q->nr)
goto free_queue;
- if (output_format & (DIFF_FORMAT_RAW |
- DIFF_FORMAT_NAME |
+ if ((output_format & (DIFF_FORMAT_RAW |
DIFF_FORMAT_NAME_STATUS |
- DIFF_FORMAT_CHECKDIFF)) {
+ DIFF_FORMAT_CHECKDIFF)) ||
+ (output_format & DIFF_FORMAT_NAME && !options->flags.diff_from_contents)) {
for (i = 0; i < q->nr; i++) {
struct diff_filepair *p = q->queue[i];
if (check_pair_status(p))
@@ -6449,9 +6451,10 @@ void diff_flush(struct diff_options *options)
separator++;
}
- if (output_format & DIFF_FORMAT_NO_OUTPUT &&
+ if ((output_format & DIFF_FORMAT_NO_OUTPUT &&
options->flags.exit_with_status &&
- options->flags.diff_from_contents) {
+ options->flags.diff_from_contents)) ||
+ (output_format & DIFF_FORMAT_NAME && options->flags.diff_from_contents) {
/*
* run diff_flush_patch for the exit status. setting
* options->file to /dev/null should be safe, because we