threads / discuss / 25365

git diff --check always shows line 1 for blank at EOF for new files

Subject: git diff --check always shows line 1 for blank at EOF for new files

## tl;dr

4 messages between Oct 6, 2010 and Oct 10, 2010.

replies: 3people: 3as markdown or json

Christoph Mallon· Oct 6, 2010, 09:27 UTC · lore
Hi,

there seems to be a glitch in git diff --check for new files: If a new file has trailing empty lines, then the error message always refers to line 1. This happens with git 1.7.3.1.

Here is a simple test case:
   git init test
   cd test
   printf 'a\nb\nc\n\n' > file
   git add -AN
   git diff --check
The last command will show "file:1: new blank line at EOF.".
It works fine, if the diff is not against /dev/null, e.g.:
   git init test
   cd test
   printf 'a\nb\nc\n' > file
   git add -A
   echo >> file
   git diff --check
This correctly shows "file:4: new blank line at EOF.".
Jonathan Nieder· Oct 7, 2010, 09:32 UTC · re: Christoph Mallon · lore

Re: git diff --check always shows line 1 for blank at EOF for new files

Hi Christoph,
Christoph Mallon wrote:
Show 17 quoted lines
>   git init test
>   cd test
>   printf 'a\nb\nc\n\n' > file
>   git add -AN
>   git diff --check
> 
> The last command will show "file:1: new blank line at EOF.".
> 
> It works fine, if the diff is not against /dev/null, e.g.:
>   git init test
>   cd test
>   printf 'a\nb\nc\n' > file
>   git add -A
>   echo >> file
>   git diff --check
> 
> This correctly shows "file:4: new blank line at EOF.".
Try this:
    git init test
    cd test
    printf 'a\nb\nc\n' > file
    git add -A
    echo hello >> file
    echo >> file
    git diff --check

As you can see, it still returns line 4, which is the beginning of the hunk that adds the blank line.

A change to make it print the line number of the blank line itself does not sound so bad, though.

Christoph Mallon· Oct 7, 2010, 12:33 UTC · re: Jonathan Nieder · lore

Re: git diff --check always shows line 1 for blank at EOF for new files

On 07.10.2010 11:32, Jonathan Nieder wrote:
Show 15 quoted lines
> Try this:
>
>      git init test
>      cd test
>      printf 'a\nb\nc\n'>  file
>      git add -A
>      echo hello>>  file
>      echo>>  file
>      git diff --check
>
> As you can see, it still returns line 4, which is the beginning of the
> hunk that adds the blank line.
>
> A change to make it print the line number of the blank line itself
> does not sound so bad, though.
You're right. I think, the attached patch corrects the issue.
>From 45eaa882dae35f2976e77cf0b6b06be78283c13f Mon Sep 17 00:00:00 2001
From: Christoph Mallon <christoph.mallon@gmx.de>
Date: Thu, 7 Oct 2010 14:22:02 +0200
Subject: [PATCH] When git diff --check detects a blank line at EOF, show the line number of the empty line, not the line number of the beginning of the hunk.
---
 diff.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a/diff.c b/diff.c
index 71efa8e..452fdf4 100644
--- a/diff.c
+++ b/diff.c
@@ -2153,7 +2153,7 @@ static void builtin_checkdiff(const char *name_a, const char *name_b,
 
 			ecbdata.ws_rule = data.ws_rule;
 			check_blank_at_eof(&mf1, &mf2, &ecbdata);
-			blank_at_eof = ecbdata.blank_at_eof_in_preimage;
+			blank_at_eof = ecbdata.blank_at_eof_in_postimage;
 
 			if (blank_at_eof) {
 				static char *err;

-- 
1.7.3
Junio C Hamano· Oct 10, 2010, 10:46 UTC · re: Christoph Mallon · lore

Re: git diff --check always shows line 1 for blank at EOF for new files

Christoph Mallon <christoph.mallon@gmx.de> writes:
Show 25 quoted lines
> On 07.10.2010 11:32, Jonathan Nieder wrote:
>> Try this:
>>
>>      git init test
>>      cd test
>>      printf 'a\nb\nc\n'>  file
>>      git add -A
>>      echo hello>>  file
>>      echo>>  file
>>      git diff --check
>>
>> As you can see, it still returns line 4, which is the beginning of the
>> hunk that adds the blank line.
>>
>> A change to make it print the line number of the blank line itself
>> does not sound so bad, though.
>
> You're right. I think, the attached patch corrects the issue.
>
> From 45eaa882dae35f2976e77cf0b6b06be78283c13f Mon Sep 17 00:00:00 2001
> From: Christoph Mallon <christoph.mallon@gmx.de>
> Date: Thu, 7 Oct 2010 14:22:02 +0200
> Subject: [PATCH] When git diff --check detects a blank line at EOF, show the line number of the empty line, not the line number of the beginning of the hunk.
>
> ---
Could you check Documentation/SubmittingPatches, please?
Show 13 quoted lines
>  diff.c |    2 +-
>  1 files changed, 1 insertions(+), 1 deletions(-)
>
> diff --git a/diff.c b/diff.c
> index 71efa8e..452fdf4 100644
> --- a/diff.c
> +++ b/diff.c
> @@ -2153,7 +2153,7 @@ static void builtin_checkdiff(const char *name_a, const char *name_b,
>  
>  			ecbdata.ws_rule = data.ws_rule;
>  			check_blank_at_eof(&mf1, &mf2, &ecbdata);
> -			blank_at_eof = ecbdata.blank_at_eof_in_preimage;
> +			blank_at_eof = ecbdata.blank_at_eof_in_postimage;
Hmm, I am confused.  Your subject line claims that for this patch:
        diff --git a/Makefile b/Makefile
        index 1f1ce04..90075e0 100644
        --- a/Makefile
        +++ b/Makefile
        @@ -2375,3 +2375,8 @@ cover_db: coverage-report
         cover_db_html: cover_db
                cover -report html -outputdir cover_db_html cover_db
        +
        +new-rule:
        +	echo here is a new rule
        +
        +

the code used to diagnose the error by showing the first line of the hunk (2375 taken from either -2375 or +2375, it does not matter much). The claim is that your patch makes it to show the first line in the trailing blank line block (the line after "here is a new rule", i.e. line 2381 in the postimage).

But my reading of the code is that it is showing (incorrectly) the first line of the trailing blank line block in the preimage (in this case, because our codebase is clean, there is zero line of trailing blank line block in the preimage, so it points at 2378.

So the patch looks correct, but the diagnosis / description seems wrong. Since I am asking you to refer to SubmittingPatches to give a better subject line, patch description and a sign-off anyway,...

Thanks.

← back to recent threads