[PATCH/docs] make slash-rules more readable

[PATCH/docs] make slash-rules more readable

[Dr. Adam Nielsen]

From: Adam Nielsen <admin@in-ici.net>

gitignore.txt: make slash-rules more readable

Remove the addition `it is removed for the purpose of the following description` and 
make clear in which situations a trailing slash is used or not. Increase readability
and make all paragraphs valid, even if they are not read in strict order.
Replace `otherwise` with the the concrete pattern that is considered in the paragraph to avoid
confusion. 
Add simple examples to point out the significant difference between using or not using a trailing slash.

Signed-off-by: Adam J. N. Nielsen <info@drnielsen.de>

---
 Documentation/gitignore.txt | 23 +++++++++++++----------
 1 file changed, 13 insertions(+), 10 deletions(-)

diff --git a/Documentation/gitignore.txt b/Documentation/gitignore.txt
index 1c94f08ff4..c6720b0ac4 100644
--- a/Documentation/gitignore.txt
+++ b/Documentation/gitignore.txt
@@ -89,22 +89,25 @@ PATTERN FORMAT
    Put a backslash ("`\`") in front of the first "`!`" for patterns
    that begin with a literal "`!`", for example, "`\!important!.txt`".
 
- - If the pattern ends with a slash, it is removed for the
-   purpose of the following description, but it would only find
+ - If the pattern ends with a slash "`/`", it would only find
    a match with a directory.  In other words, `foo/` will match a
    directory `foo` and paths underneath it, but will not match a
    regular file or a symbolic link `foo` (this is consistent
    with the way how pathspec works in general in Git).
 
- - If the pattern does not contain a slash '/', Git treats it as
-   a shell glob pattern and checks for a match against the
-   pathname relative to the location of the `.gitignore` file
-   (relative to the toplevel of the work tree if not from a
-   `.gitignore` file).
+ - If the pattern contains no slash "`/`" other then a trailing slash,
+   then the pattern will match in all directories. In other words,
+   `foo/` will match `/bar/foo/` and `foo` will match `/bar/bar/foo`.
 
- - Otherwise, Git treats the pattern as a shell glob: "`*`" matches
-   anything except "`/`", "`?`" matches any one character except "`/`"
-   and "`[]`" matches one character in a selected range. See
+ - If the pattern contains a slash "`/`" other then a trailing slash, then
+   the pattern is always considered from the `.gitignore` file location.
+   In other words, `foo/bar` will match `/foo/bar` but not `/bar/foo/bar`.
+
+ - The character "`*`" matches anything except a non trailing slash "`/`".
+   For example, "foo/*" matches "foo/test.json" and "foo/bar/"
+   but not "foo/bar/test.json".
+   The character "`?`" matches any one character except "`/`".
+   The character "`[]`" matches one character in a selected range. See
    fnmatch(3) and the FNM_PATHNAME flag for a more detailed
    description.
 
-- 
2.17.1

Re: [PATCH/docs] make slash-rules more readable

[Junio C Hamano]

"Dr. Adam Nielsen" <admin@in-ici.net> writes:

A few notes on the form.

> From: Adam Nielsen <admin@in-ici.net>

This "author" identity and the name-email on the Signed-off-by: line
should match, at least for this project.  I cannot tell which one is
your preference, and I do not have any preference over your name
either ;-), but please pick one and use it consistently.

>
> gitignore.txt: make slash-rules more readable
>
> Remove the addition `it is removed for the purpose of the following description` and 
> make clear in which situations a trailing slash is used or not. Increase readability
> and make all paragraphs valid, even if they are not read in strict order.
> Replace `otherwise` with the the concrete pattern that is considered in the paragraph to avoid
> confusion. 
> Add simple examples to point out the significant difference between using or not using a trailing slash.

These are overly long lines; we tend to fold long lines at around
70 char or so.

> Signed-off-by: Adam J. N. Nielsen <info@drnielsen.de>
>
> ---
>  Documentation/gitignore.txt | 23 +++++++++++++----------
>  1 file changed, 13 insertions(+), 10 deletions(-)
>
> diff --git a/Documentation/gitignore.txt b/Documentation/gitignore.txt
> index 1c94f08ff4..c6720b0ac4 100644
> --- a/Documentation/gitignore.txt
> +++ b/Documentation/gitignore.txt
> @@ -89,22 +89,25 @@ PATTERN FORMAT
>     Put a backslash ("`\`") in front of the first "`!`" for patterns
>     that begin with a literal "`!`", for example, "`\!important!.txt`".
>  
> - - If the pattern ends with a slash, it is removed for the
> -   purpose of the following description, but it would only find
> + - If the pattern ends with a slash "`/`", it would only find
>     a match with a directory.  In other words, `foo/` will match a
>     directory `foo` and paths underneath it, but will not match a
>     regular file or a symbolic link `foo` (this is consistent
>     with the way how pathspec works in general in Git).

I do like this change, even though I cannot bring myself backing it
100% immediately.  The reason why I wrote it the way in the original
was because I did not want to repeat "... but a slash at end, if
exists, is exempt from this rule" over and over in the later bullet
points, as it would be a maintenance burden when we have more bullet
points and when we find a better phrasing to say "... but a slash at
end if exists, is exempt from this rule".

The patch I am responding to bites the bullet and repeats the "the
one at the end does not count", which may be slightly harder to
maintain, but certainly makes it easier to read.

> - - If the pattern does not contain a slash '/', Git treats it as
> -   a shell glob pattern and checks for a match against the
> -   pathname relative to the location of the `.gitignore` file
> -   (relative to the toplevel of the work tree if not from a
> -   `.gitignore` file).
> + - If the pattern contains no slash "`/`" other then a trailing slash,

While pretending to be a fresh reader and reading only this line
made me wonder if the rule described in this bullet point applies
only to a pattern that has a single slash at the end.  I wonder if
it is just me, or we can improve the phrasing so that it is clear
that a pattern without any slash also is covered by this rule, not
just a pattern that has all non-slash chars followed by a single
slash.

> +   then the pattern will match in all directories. In other words,
> +   `foo/` will match `/bar/foo/` and `foo` will match `/bar/bar/foo`.

The half-technical "treats it as a shell glob pattern" from the
original is gone, which I think is a good change.  The examples may
need to be improved, as it may not be clear to naive readers that
with /bar/foo/, you meant that it is limited to a directory but not
a file, and with /bar/bar/foo you meant both a directory and a file
is fine.  Perhaps

	For example, 'frotz/' matches 'frotz', 'a/frotz', etc. that
	is a directory, but does not match if these are files.
	A pattern 'frotz' on the other hand matches these paths
	whether they are files or directories.

I also wonder if "in all directories" is clear enough that your
"all" is limited to below the level the ignore pattern is defined
for (i.e. "*.1" that appears in "Documentation/.gitignore" does not
ignore "foo.1" at the top-level of the tree).

So I can tell that this patch is trying to address a problem in the
original that is worth fixing, but I cannot say the result is good.
At least not yet.

> - - Otherwise, Git treats the pattern as a shell glob: "`*`" matches
> -   anything except "`/`", "`?`" matches any one character except "`/`"
> -   and "`[]`" matches one character in a selected range. See
> + - If the pattern contains a slash "`/`" other then a trailing slash, then

The same comment applies to this first line about the ambiguity of a
pattern without any slash anywhere.

> +   the pattern is always considered from the `.gitignore` file location.
> +   In other words, `foo/bar` will match `/foo/bar` but not `/bar/foo/bar`.

Again, loss of the mention of "shell glob" is a good thing, as we
still have a clue for those "in the know" at the end by mentioning
fnmatch(3).

The example lacks one crucial description to be useful.  The reader
must be told where foo/bar came from.  Was it in the .gitignore file
at the top-level?  A per-directory exclude file bar/.gitignore?
Without making that clear, none of the "In other words" example
makes much sense.

Also another issue common to previous example is that you are using
absolute path notation "/bar/foo/", "/bar/foo/bar", etc. without
explaining what you want it mean.  I can guess that it does not
refer to the root of the filesystem but you meant to refer to the
top level of the working tree, but you are not writing documentation
to help _me_ understand Git, so we should not rely on that "I can
guess".  I do not think an average first-time reader can.

> + - The character "`*`" matches anything except a non trailing slash "`/`".
> +   For example, "foo/*" matches "foo/test.json" and "foo/bar/"
> +   but not "foo/bar/test.json".

I think your writing out the trailing slash on the filesystem-entity
side (i.e. things that are matched by patterns) is making the
resulting description more distracting than necessary.  Being able
to mark a pattern with a trailing slash to "match only to directory"
is one thing, but when the example talks about paths foo/test.json
(presumably a regular file and not a directory) and foo/bar
(presumably a directory), it shouldn't force users to mistakenly
think that the matching engine first appends a slash after a
directory we read from the filesystem before applying the pattern
matching logic, which has a compensating hack to ignore trailing
slash from the path when matching.

Once you write consistently that a path for a directory foo/bar is
foo/bar, not foo/bar/, then this example would become much easier to
write and read, I suspect.

	An asterisk "`*`" matches anything except a slash.  A
	pattern "foo/*", for example, matches "foo/test.json" (a
	regular file), "foo/bar" (a diretory), but it does not match
	"foo/bar/hello.c" (a regular file), as the asterisk in the
	patter does not match "bar/hello.c" which has a slash in it.

perhaps.

> +   The character "`?`" matches any one character except "`/`".
> +   The character "`[]`" matches one character in a selected range. See

Calling `[]` construct "the character" is blatantly wrong.

	The range notation, e.g. `[a-zA-Z]`, can be used to match
	one of the characters in a range.

perhaps.  It still omits negation [!0-9] but it probably is OK to
leave that for fnmatch(3), and you've done so by leaving these two
lines from the original intact, which is good.

>     fnmatch(3) and the FNM_PATHNAME flag for a more detailed
>     description.


Thanks.

Re: [PATCH/docs] make slash-rules more readable

[Dr. Adam Nielsen]

Am Mo., 8. Apr. 2019 um 09:51 Uhr schrieb Junio C Hamano <gitster@pobox.com>:
>
> "Adam Nielsen" <admin@in-ici.net> writes:
>
> A few notes on the form.
>
> > From: Adam Nielsen <admin@in-ici.net>
>
> This "author" identity and the name-email on the Signed-off-by: line
> should match, at least for this project.  I cannot tell which one is
> your preference, and I do not have any preference over your name
> either ;-), but please pick one and use it consistently.
>

Haha yes sorry. I had my struggles with this patch procedure and I
will do better next time.


> >
> > gitignore.txt: make slash-rules more readable
> >
> > Remove the addition `it is removed for the purpose of the following description` and
> > make clear in which situations a trailing slash is used or not. Increase readability
> > and make all paragraphs valid, even if they are not read in strict order.
> > Replace `otherwise` with the the concrete pattern that is considered in the paragraph to avoid
> > confusion.
> > Add simple examples to point out the significant difference between using or not using a trailing slash.
>
> These are overly long lines; we tend to fold long lines at around
> 70 char or so.

Okay.

>
> > Signed-off-by: Adam J. N. Nielsen <info@drnielsen.de>
> >
> > ---
> >  Documentation/gitignore.txt | 23 +++++++++++++----------
> >  1 file changed, 13 insertions(+), 10 deletions(-)
> >
> > diff --git a/Documentation/gitignore.txt b/Documentation/gitignore.txt
> > index 1c94f08ff4..c6720b0ac4 100644
> > --- a/Documentation/gitignore.txt
> > +++ b/Documentation/gitignore.txt
> > @@ -89,22 +89,25 @@ PATTERN FORMAT
> >     Put a backslash ("`\`") in front of the first "`!`" for patterns
> >     that begin with a literal "`!`", for example, "`\!important!.txt`".
> >
> > - - If the pattern ends with a slash, it is removed for the
> > -   purpose of the following description, but it would only find
> > + - If the pattern ends with a slash "`/`", it would only find
> >     a match with a directory.  In other words, `foo/` will match a
> >     directory `foo` and paths underneath it, but will not match a
> >     regular file or a symbolic link `foo` (this is consistent
> >     with the way how pathspec works in general in Git).
>
> I do like this change, even though I cannot bring myself backing it
> 100% immediately.

> The reason why I wrote it the way in the original
> was because I did not want to repeat "... but a slash at end, if
> exists, is exempt from this rule"

Yes, I can see why this makes sense. However, I find that this exception
makes this paragraph hard to read. Also I think its ambiguous if "it" refers
to the pattern or the slash. The first few times I read it,
I just didn't get it and was very intimidated by the paragraph.

> over and over in the later bullet
> points, as it would be a maintenance burden when we have more bullet
> points and when we find a better phrasing to say "... but a slash at
> end if exists, is exempt from this rule".

Yes, I agree. One should not repeat such a bloated exception rule over and over
again.

>
> The patch I am responding to bites the bullet and repeats the "the
> one at the end does not count", which may be slightly harder to
> maintain, but certainly makes it easier to read.
>
> > - - If the pattern does not contain a slash '/', Git treats it as
> > -   a shell glob pattern and checks for a match against the
> > -   pathname relative to the location of the `.gitignore` file
> > -   (relative to the toplevel of the work tree if not from a
> > -   `.gitignore` file).
> > + - If the pattern contains no slash "`/`" other then a trailing slash,
>
> While pretending to be a fresh reader and reading only this line
> made me wonder if the rule described in this bullet point applies
> only to a pattern that has a single slash at the end.  I wonder if
> it is just me, or we can improve the phrasing so that it is clear
> that a pattern without any slash also is covered by this rule, not
> just a pattern that has all non-slash chars followed by a single
> slash.

I agree with you. How about we make up the word "intermediate slash" and
explain it in an extra paragraph? This would make it less repetitive.
Also it makes clear that the case without any slash is also covered.
Perhaps

         In the following we use the term **intermediate slash** to
         denote a slash "`/`" in a pattern that is not a trailing slash
         nor a leading slash.
         For example the pattern `/foo/bar`, `foo/bar` and `/foo/bar/`
all contain
         only one intermediate slash. The pattern `foo/` does not contain an
         intermediate slash.

Then, instead of:

         If the pattern contains no slash "`/`" other then a trailing slash,"

one could say:

         If the pattern contains no intermediate slash "`/`",


>
> > +   then the pattern will match in all directories. In other words,
> > +   `foo/` will match `/bar/foo/` and `foo` will match `/bar/bar/foo`.
>
> The half-technical "treats it as a shell glob pattern" from the
> original is gone, which I think is a good change.  The examples may
> need to be improved, as it may not be clear to naive readers that
> with /bar/foo/, you meant that it is limited to a directory but not
> a file, and with /bar/bar/foo you meant both a directory and a file
> is fine.  Perhaps
>
>         For example, 'frotz/' matches 'frotz', 'a/frotz', etc. that
>         is a directory, but does not match if these are files.
>         A pattern 'frotz' on the other hand matches these paths
>         whether they are files or directories.
>

Yes. This is so much better.

> I also wonder if "in all directories" is clear enough that your
> "all" is limited to below the level the ignore pattern is defined
> for (i.e. "*.1" that appears in "Documentation/.gitignore" does not
> ignore "foo.1" at the top-level of the tree).

Its mentioned at the start of the page that the pattern is always
relative to the location of the `.gitignore` file. However, I see that
since its said "in all directories" its necessary to restrict it again.
How about

         If the pattern contains no intermediate slash "`/`",
         the pattern will match in all directories at or below
         the `.gitignore` file, with infinite depth.

>
> So I can tell that this patch is trying to address a problem in the
> original that is worth fixing, but I cannot say the result is good.
> At least not yet.
>
> > - - Otherwise, Git treats the pattern as a shell glob: "`*`" matches
> > -   anything except "`/`", "`?`" matches any one character except "`/`"
> > -   and "`[]`" matches one character in a selected range. See
> > + - If the pattern contains a slash "`/`" other then a trailing slash, then
>
> The same comment applies to this first line about the ambiguity of a
> pattern without any slash anywhere.

This would now change with the above remarks to:

         If the pattern contains an intermediate slash "`/`",

>
> > +   the pattern is always considered from the `.gitignore` file location.
> > +   In other words, `foo/bar` will match `/foo/bar` but not `/bar/foo/bar`.
>
> Again, loss of the mention of "shell glob" is a good thing, as we
> still have a clue for those "in the know" at the end by mentioning
> fnmatch(3).
>
> The example lacks one crucial description to be useful.  The reader
> must be told where foo/bar came from.  Was it in the .gitignore file
> at the top-level?  A per-directory exclude file bar/.gitignore?
> Without making that clear, none of the "In other words" example
> makes much sense.
>
> Also another issue common to previous example is that you are using
> absolute path notation "/bar/foo/", "/bar/foo/bar", etc. without
> explaining what you want it mean.  I can guess that it does not
> refer to the root of the filesystem but you meant to refer to the
> top level of the working tree, but you are not writing documentation
> to help _me_ understand Git, so we should not rely on that "I can
> guess".  I do not think an average first-time reader can.

Maybe its shorter and clearer to write it like this:

         If the pattern contains an intermediate slash "`/`",
         its equivalent to the same pattern starting with a leading slash.
         For example the pattern `doc/read.txt` is equivalent to
         `/doc/read.txt`.

If we do this, one would need to lift the "leading slash" paragraph up
(The one starting with "A leading slash matches the beginning of the
pathname. For example,...").

Note that since an intermediate slash is explicitly not a leading
slash, it is not said that
`/bar/` and `//bar/` are equivalent.



>
> > + - The character "`*`" matches anything except a non trailing slash "`/`".
> > +   For example, "foo/*" matches "foo/test.json" and "foo/bar/"
> > +   but not "foo/bar/test.json".
>
> I think your writing out the trailing slash on the filesystem-entity
> side (i.e. things that are matched by patterns) is making the
> resulting description more distracting than necessary.  Being able
> to mark a pattern with a trailing slash to "match only to directory"
> is one thing, but when the example talks about paths foo/test.json
> (presumably a regular file and not a directory) and foo/bar
> (presumably a directory), it shouldn't force users to mistakenly
> think that the matching engine first appends a slash after a
> directory we read from the filesystem before applying the pattern
> matching logic, which has a compensating hack to ignore trailing
> slash from the path when matching.
>
> Once you write consistently that a path for a directory foo/bar is
> foo/bar, not foo/bar/, then this example would become much easier to
> write and read, I suspect.
>
>         An asterisk "`*`" matches anything except a slash.  A
>         pattern "foo/*", for example, matches "foo/test.json" (a
>         regular file), "foo/bar" (a diretory), but it does not match
>         "foo/bar/hello.c" (a regular file), as the asterisk in the
>         patter does not match "bar/hello.c" which has a slash in it.
>
> perhaps.

I agree, this is much better. Although I would leave out

>  "as the asterisk in the patter does not match "bar/hello.c"
>   which has a slash in it."

>
> > +   The character "`?`" matches any one character except "`/`".
> > +   The character "`[]`" matches one character in a selected range. See
>
> Calling `[]` construct "the character" is blatantly wrong.

Yes, my bad.

>
>         The range notation, e.g. `[a-zA-Z]`, can be used to match
>         one of the characters in a range.
>
> perhaps.

That is much better too.

> It still omits negation [!0-9] but it probably is OK to
> leave that for fnmatch(3), and you've done so by leaving these two
> lines from the original intact, which is good.
>
> >     fnmatch(3) and the FNM_PATHNAME flag for a more detailed
> >     description.
>
>
> Thanks.

Thank you for all your input. If you agree with my proposed changes, I would
create a new patch merging all this together.


Am Mo., 8. Apr. 2019 um 09:51 Uhr schrieb Junio C Hamano <gitster@pobox.com>:
>
> "Dr. Adam Nielsen" <admin@in-ici.net> writes:
>
> A few notes on the form.
>
> > From: Adam Nielsen <admin@in-ici.net>
>
> This "author" identity and the name-email on the Signed-off-by: line
> should match, at least for this project.  I cannot tell which one is
> your preference, and I do not have any preference over your name
> either ;-), but please pick one and use it consistently.
>
> >
> > gitignore.txt: make slash-rules more readable
> >
> > Remove the addition `it is removed for the purpose of the following description` and
> > make clear in which situations a trailing slash is used or not. Increase readability
> > and make all paragraphs valid, even if they are not read in strict order.
> > Replace `otherwise` with the the concrete pattern that is considered in the paragraph to avoid
> > confusion.
> > Add simple examples to point out the significant difference between using or not using a trailing slash.
>
> These are overly long lines; we tend to fold long lines at around
> 70 char or so.
>
> > Signed-off-by: Adam J. N. Nielsen <info@drnielsen.de>
> >
> > ---
> >  Documentation/gitignore.txt | 23 +++++++++++++----------
> >  1 file changed, 13 insertions(+), 10 deletions(-)
> >
> > diff --git a/Documentation/gitignore.txt b/Documentation/gitignore.txt
> > index 1c94f08ff4..c6720b0ac4 100644
> > --- a/Documentation/gitignore.txt
> > +++ b/Documentation/gitignore.txt
> > @@ -89,22 +89,25 @@ PATTERN FORMAT
> >     Put a backslash ("`\`") in front of the first "`!`" for patterns
> >     that begin with a literal "`!`", for example, "`\!important!.txt`".
> >
> > - - If the pattern ends with a slash, it is removed for the
> > -   purpose of the following description, but it would only find
> > + - If the pattern ends with a slash "`/`", it would only find
> >     a match with a directory.  In other words, `foo/` will match a
> >     directory `foo` and paths underneath it, but will not match a
> >     regular file or a symbolic link `foo` (this is consistent
> >     with the way how pathspec works in general in Git).
>
> I do like this change, even though I cannot bring myself backing it
> 100% immediately.  The reason why I wrote it the way in the original
> was because I did not want to repeat "... but a slash at end, if
> exists, is exempt from this rule" over and over in the later bullet
> points, as it would be a maintenance burden when we have more bullet
> points and when we find a better phrasing to say "... but a slash at
> end if exists, is exempt from this rule".
>
> The patch I am responding to bites the bullet and repeats the "the
> one at the end does not count", which may be slightly harder to
> maintain, but certainly makes it easier to read.
>
> > - - If the pattern does not contain a slash '/', Git treats it as
> > -   a shell glob pattern and checks for a match against the
> > -   pathname relative to the location of the `.gitignore` file
> > -   (relative to the toplevel of the work tree if not from a
> > -   `.gitignore` file).
> > + - If the pattern contains no slash "`/`" other then a trailing slash,
>
> While pretending to be a fresh reader and reading only this line
> made me wonder if the rule described in this bullet point applies
> only to a pattern that has a single slash at the end.  I wonder if
> it is just me, or we can improve the phrasing so that it is clear
> that a pattern without any slash also is covered by this rule, not
> just a pattern that has all non-slash chars followed by a single
> slash.
>
> > +   then the pattern will match in all directories. In other words,
> > +   `foo/` will match `/bar/foo/` and `foo` will match `/bar/bar/foo`.
>
> The half-technical "treats it as a shell glob pattern" from the
> original is gone, which I think is a good change.  The examples may
> need to be improved, as it may not be clear to naive readers that
> with /bar/foo/, you meant that it is limited to a directory but not
> a file, and with /bar/bar/foo you meant both a directory and a file
> is fine.  Perhaps
>
>         For example, 'frotz/' matches 'frotz', 'a/frotz', etc. that
>         is a directory, but does not match if these are files.
>         A pattern 'frotz' on the other hand matches these paths
>         whether they are files or directories.
>
> I also wonder if "in all directories" is clear enough that your
> "all" is limited to below the level the ignore pattern is defined
> for (i.e. "*.1" that appears in "Documentation/.gitignore" does not
> ignore "foo.1" at the top-level of the tree).
>
> So I can tell that this patch is trying to address a problem in the
> original that is worth fixing, but I cannot say the result is good.
> At least not yet.
>
> > - - Otherwise, Git treats the pattern as a shell glob: "`*`" matches
> > -   anything except "`/`", "`?`" matches any one character except "`/`"
> > -   and "`[]`" matches one character in a selected range. See
> > + - If the pattern contains a slash "`/`" other then a trailing slash, then
>
> The same comment applies to this first line about the ambiguity of a
> pattern without any slash anywhere.
>
> > +   the pattern is always considered from the `.gitignore` file location.
> > +   In other words, `foo/bar` will match `/foo/bar` but not `/bar/foo/bar`.
>
> Again, loss of the mention of "shell glob" is a good thing, as we
> still have a clue for those "in the know" at the end by mentioning
> fnmatch(3).
>
> The example lacks one crucial description to be useful.  The reader
> must be told where foo/bar came from.  Was it in the .gitignore file
> at the top-level?  A per-directory exclude file bar/.gitignore?
> Without making that clear, none of the "In other words" example
> makes much sense.
>
> Also another issue common to previous example is that you are using
> absolute path notation "/bar/foo/", "/bar/foo/bar", etc. without
> explaining what you want it mean.  I can guess that it does not
> refer to the root of the filesystem but you meant to refer to the
> top level of the working tree, but you are not writing documentation
> to help _me_ understand Git, so we should not rely on that "I can
> guess".  I do not think an average first-time reader can.
>
> > + - The character "`*`" matches anything except a non trailing slash "`/`".
> > +   For example, "foo/*" matches "foo/test.json" and "foo/bar/"
> > +   but not "foo/bar/test.json".
>
> I think your writing out the trailing slash on the filesystem-entity
> side (i.e. things that are matched by patterns) is making the
> resulting description more distracting than necessary.  Being able
> to mark a pattern with a trailing slash to "match only to directory"
> is one thing, but when the example talks about paths foo/test.json
> (presumably a regular file and not a directory) and foo/bar
> (presumably a directory), it shouldn't force users to mistakenly
> think that the matching engine first appends a slash after a
> directory we read from the filesystem before applying the pattern
> matching logic, which has a compensating hack to ignore trailing
> slash from the path when matching.
>
> Once you write consistently that a path for a directory foo/bar is
> foo/bar, not foo/bar/, then this example would become much easier to
> write and read, I suspect.
>
>         An asterisk "`*`" matches anything except a slash.  A
>         pattern "foo/*", for example, matches "foo/test.json" (a
>         regular file), "foo/bar" (a diretory), but it does not match
>         "foo/bar/hello.c" (a regular file), as the asterisk in the
>         patter does not match "bar/hello.c" which has a slash in it.
>
> perhaps.
>
> > +   The character "`?`" matches any one character except "`/`".
> > +   The character "`[]`" matches one character in a selected range. See
>
> Calling `[]` construct "the character" is blatantly wrong.
>
>         The range notation, e.g. `[a-zA-Z]`, can be used to match
>         one of the characters in a range.
>
> perhaps.  It still omits negation [!0-9] but it probably is OK to
> leave that for fnmatch(3), and you've done so by leaving these two
> lines from the original intact, which is good.
>
> >     fnmatch(3) and the FNM_PATHNAME flag for a more detailed
> >     description.
>
>
> Thanks.



-- 


Dr. Adam Nielsen


Administrator for IN/ICI/WHO


IN:

www.nlp-institutes.net


ICI:

www.coaching-institutes.net


WHO:

www.world-hypnosis.org

Re: [PATCH/docs] make slash-rules more readable

[Junio C Hamano]

"Dr. Adam Nielsen" <admin@in-ici.net> writes:

> I agree with you. How about we make up the word "intermediate slash" and
> explain it in an extra paragraph?

I am not sure if that is any better than "in the following, pretend
that a slash at the end of a pattern does not exist", which is how
the current description avoids repetition and aims for clarity.  It
probably is worse than than the current one if we need to introduce
a new term that is otherwise not useful elsewhere---a new term adds
to the cognitive load of readers.

>> I also wonder if "in all directories" is clear enough that your
>> "all" is limited to below the level the ignore pattern is defined
>> for (i.e. "*.1" that appears in "Documentation/.gitignore" does not
>> ignore "foo.1" at the top-level of the tree).
>
> Its mentioned at the start of the page that the pattern is always
> relative to the location of the `.gitignore` file. However, I see that
> since its said "in all directories" its necessary to restrict it again.
> How about
>
>          If the pattern contains no intermediate slash "`/`",
>          the pattern will match in all directories at or below
>          the `.gitignore` file, with infinite depth.

It is unclear what "with infinite depth" means in this sentence.
There is no depth-limit in the exclude mechanism, and I'd prefer
not to confuse readers by making a casual mention of "depth" to
imply as if there is some depth-based logic.

Also, as you defined "intermediate" as a slash that is neither
leading nor trailing, the above paragraph says "/foo" matches any
filesystem entity whose final path component is 'foo', e.g. a file
'foo' at the current level, a directory 'foo' in subdirectory 'dir'
(i.e. 'dir/foo'), etc.  I do not think you meant to say that (and
this is why I do not like to introduce a new term---even its
inventor cannot get it right).

>> So I can tell that this patch is trying to address a problem in the
>> original that is worth fixing, but I cannot say the result is good.
>> At least not yet.
>> ...
>> Once you write consistently that a path for a directory foo/bar is
>> foo/bar, not foo/bar/, then this example would become much easier to
>> write and read, I suspect.
>>
>>         An asterisk "`*`" matches anything except a slash.  A
>>         pattern "foo/*", for example, matches "foo/test.json" (a
>>         regular file), "foo/bar" (a diretory), but it does not match
>>         "foo/bar/hello.c" (a regular file), as the asterisk in the
>>         patter does not match "bar/hello.c" which has a slash in it.
>>
>> perhaps.
>
> I agree, this is much better. Although I would leave out
>
>>  "as the asterisk in the patter does not match "bar/hello.c"
>>   which has a slash in it."

I happen to think that the part is the more important half of that
whole "example".  By explaining why it does not match, it enforces
"matches anything except a slash" we gave upfront.

Thanks.  I think we are making progress...

Re: [PATCH/docs] make slash-rules more readable

[Dr. Adam Nielsen]

>> I agree with you. How about we make up the word "intermediate slash" and
>> explain it in an extra paragraph?

> I am not sure if that is any better than "in the following, pretend
> that a slash at the end of a pattern does not exist", which is how
> the current description avoids repetition and aims for clarity.
> It probably is worse than than the current one if we need to introduce
> a new term that is otherwise not useful elsewhere---a new term adds
> to the cognitive load of readers.

The description "in the following, pretend that a slash at the end of
a pattern does not exist"
is much better understandable then "it is removed for the purpose of
the following description"
to me. However, I still think its better not to use it:

  1. The following paragraphs may not be valid if you do not keep this
rule in mind.
      This could happen, because one may not read the paragraphs in order,
      or its unclear if "in the following" is meant for all following
paragraphs or just this
      paragraph, or one simply forgets it if one happens to check up a
specific paragraph
      after some months or when returning from a break.
  2. Forcing the user to be aware that "in the following pretend that
a slash at the end does not exist"
      is at least for me a harder cognitive load then introducing a new term.
      Because for every new paragraph that I read, I have to check if
this rule applies.
  3. If you jump between paragraphs, you always need to locate the rule
      and then check if the current paragraph is below or after that rule.
  4. There are 7 paragraphs following this "specific rule" but only
the next paragraph
      is using it. I think its also not needed for the paragraph where
it appears.
      For the other 6 paragraphs one needs to unnecessary check
      if the rule is implicitly used anywhere.
  5. For every paragraph that one may add in the future below that
rule, one needs to double check
      if its compatible with this rule. So in terms of maintenance I
think its rather a downside.

If something is cumbersome to explain and appears often, I think its
best to explain it once in detail
and refer to that instead of a bloated description. However, since
this scenario only appears in two paragraphs,
a new term is maybe not justifiable. Thus, a cumbersome but complete
description might be preferable.

> > Its mentioned at the start of the page that the pattern is always
> > relative to the location of the `.gitignore` file. However, I see that
> > since its said "in all directories" its necessary to restrict it again.
> > How about
> >
> >          If the pattern contains no intermediate slash "`/`",
> >          the pattern will match in all directories at or below
> >          the `.gitignore` file, with infinite depth.
>
> It is unclear what "with infinite depth" means in this sentence.
> There is no depth-limit in the exclude mechanism, and I'd prefer
> not to confuse readers by making a casual mention of "depth" to
> imply as if there is some depth-based logic.

The description "infinite depth" is used in the current documentation
in the paragraph
for `/**`. I thought its good to reuse something that has already been
approved by the community.
I think it points out very well that we are not only looking at
folders at the level of the `.gitignore`
file but all the folders below.

> Also, as you defined "intermediate" as a slash that is neither
> leading nor trailing, the above paragraph says "/foo" matches any
> filesystem entity whose final path component is 'foo', e.g. a file
> 'foo' at the current level, a directory 'foo' in subdirectory 'dir'
> (i.e. 'dir/foo'), etc.  I do not think you meant to say that (and

We could fix it by defining an intermediate slash as any slash that is
not a trailing one.
But then one may just call it a "non-trailing" slash which is maybe
self-explanatory enough.

> this is why I do not like to introduce a new term---even its
> inventor cannot get it right).

Especially if someone invents a new thing, I would not expect that its
perfect right from scratch.

However, I have to admit, the introduced term "intermediate" has its
flaws. Maybe using "non-trailing" or a description is better.

> > I agree, this is much better. Although I would leave out
> >
> >>  "as the asterisk in the patter does not match "bar/hello.c"
> >>   which has a slash in it."
>
> I happen to think that the part is the more important half of that
> whole "example".

Alright.



To summarize: I would suggest to drop "it is removed for the purpose
of the following description" because of the points
I made at the top and instead mention the necessary exceptions in the
two relevant paragraphs. Here my new updated proposals:

          If the pattern contains no slash "`/`"
          (except an optional trailing slash),
          the pattern will match in all directories relative to
          the `.gitignore` file, with infinite depth.
          For example, `frotz/` matches `frotz` and `a/frotz` that
          is a directory, but does not match if these are files.
          A pattern `frotz` on the other hand matches these paths
          whether they are files or directories.

I think this is better then my first proposal:

>> + - If the pattern contains no slash "`/`" other then a trailing slash,

>While pretending to be a fresh reader and reading only this line
>made me wonder if the rule described in this bullet point applies
>only to a pattern that has a single slash at the end.  I wonder if
>it is just me, or we can improve the phrasing so that it is clear
>that a pattern without any slash also is covered by this rule, not
>just a pattern that has all non-slash chars followed by a single
>slash.

The phrase "except an optional" and the example both point out that
this rule works with and without trailing slash.
I kept "infinite depth" for the reasons explained above.

For the next paragraph I would suggest this:

        If the pattern contains a non-trailing slash "`/`",
        it matches the beginning of the pathname.
        For example, the pattern `doc/frotz/` matches
        `doc/frotz` that is a directory
        but does not match `a/doc/frotz`.

Where I have taken `it matches the beginning of the pathname` from the
paragraph of a leading slash.
The example uses a trailing slash, to emphasize that this rule works
also with trailing slash, for the case
that the reader confuses the line with "If the pattern contains no
trailing slash". Alternatively I would suggest:

        If the pattern contains a slash "`/`"
        that is not a trailing slash,
        it matches the beginning of the pathname.
        For example, the pattern `doc/frotz/` matches
        `doc/frotz` that is a directory
        but does not match `a/doc/frotz`.

This might be addressing the problem of my first proposal:

>> + - If the pattern contains a slash "`/`" other then a trailing slash, then

>The same comment applies to this first line about the ambiguity of a
>pattern without any slash anywhere.



All the best,
Adam

Re: [PATCH/docs] make slash-rules more readable

[Junio C Hamano]

"Dr. Adam Nielsen" <admin@in-ici.net> writes:

>           If the pattern contains no slash "`/`"
>           (except an optional trailing slash),

That's perfect.

>           the pattern will match in all directories relative to
>           the `.gitignore` file, with infinite depth.

Maybe it is just me but "in all directories relative to the file"
does not say the same thing as

    the pattern is matched against paths in the directory where the
    `.gitignore` file that has the pattern in it is in, and any of
    its subdirectories (recursively).

which I think is what we want to say.  Especially, I do not think
the phrasing implies the match is limited to the directory and its
subdirectories, as it is unclear what the words "relative to" wanted
to mean in that sentence.

>           For example, `frotz/` matches `frotz` and `a/frotz` that
>           is a directory, but does not match if these are files.
>           A pattern `frotz` on the other hand matches these paths
>           whether they are files or directories.

Yeah this is good.

> For the next paragraph I would suggest this:
>
>         If the pattern contains a non-trailing slash "`/`",
>         it matches the beginning of the pathname.

The reference of "it" in "it matches" is fuzzy; first I was confused
as I thoguht it refers to the slash (which one?  if the pattern has
two or more slashes a/b/c).

	A pattern with a non-trailing slash matches the beginning of
	the pathname

would mean the same thing without ambiguity, but I am not sure
"matches the beginning of" is quite right.  

A pattern doc/frotz that appears in v1.0/.gitignore file would match
a file or directory v1.0/doc/frotz (pathnames in these example are
all from the top level of the repository).  

I think woe most important messages we need to deliver are

 - a pattern without slash applies in a directory the pattern
   "appears in", or any of its subdirectories (recursively).

 - a pattern with slash on the other hand is anchored at the
   directory the pattern "appears in" and does not apply to any of
   its subdirectories.

So with that in mind, perhaps

	Unlike a pattern without a slash, a pattern with a
	non-trailing slash is matched against paths immediately in
	the directory the `.gitignore` file the pattern appears in
	is stored in, and does not get used in its subdirectories..

or something?

>         For example, the pattern `doc/frotz/` matches
>         `doc/frotz` that is a directory
>         but does not match `a/doc/frotz`.

Correct, almost.  

	For example, the pattern `doc/frotz/` that appears in
	`.gitignore` at the top-level of the project matches
	`doc/frotz` directory (again, seen from the top-level), but
	not `a/doc/frotz`.

Also, a pattern "/doc" matches doc at the current level (i.e. the
directory in which .gitignore file that the pattern was taken from
is found) and not in any subdirectories.  Is that clear in the
proposed update?

Procedurally (I am writing to make sure we are on the same page for
the technical correctness, so that I can let you take care of the
ease of understanding of the end result---it is of no use if an easy
to understand document describes incorrect behaviour ;-)

 * A trailing slash is used *only* to mark that a pattern matches
   only against directories, and it is ignored by the actual pattern
   mac.thing logic.  It also is ignored by the logic to determine if
   the pattern is anchored.

   e.g. /foo-bar/ becomes /foo-bar but "must match dir" (aka "ends
   with a slash") is remembered.


 * A leading slash is used *only* to mark that a pattern is
   "anchored" at the current level, and otherwise it is ignored by
   the actual pattern matching logic.

   e.g. /foo-bar that used to be /foo-bar/ then becomes foo-bar but
   we also remember "must match here" (aka "had a slash in it").


 * A slash in a pattern that is not ignored by the above two rules
   (i.e. your "intermediate slash") marks a pattern as "anchored",
   i.e. used to match against the paths relative to the current
   level only.

   e.g. foo-bar that used to be /foo-bar that used to be /foo-bar/
   thru the above processing, as well as foo/bar that was ignored by
   the above two rules, are both anchored to the current level, and
   they do not match a/foo-bar or a/foo/bar in our immediate
   subdirectory 'a'

Re: [PATCH/docs] make slash-rules more readable

[Dr. Adam Nielsen]

> the pattern is matched against paths in the directory where the
> `.gitignore` file that has the pattern in it is in, and any of
> its subdirectories (recursively).

> the pattern will match in all directories relative to
> the `.gitignore` file, with infinite depth.

I could not catch the difference between the meaning of both.
However, I think "paths in the directory" and "directories relative to"
are maybe both ambiguous.

Since a pattern without a non-trailing slash must always be a file name or a
folder name, and does not have a leading slash, we could maybe just
say it like this:

        the pattern is matched against all files and folders (recursively)
        from the location of the `.gitignore` file.
---------


> Unlike a pattern without a slash, a pattern with a
> non-trailing slash is matched against paths immediately in
> the directory the `.gitignore` file the pattern appears in
> is stored in, and does not get used in its subdirectories..

I think one can always assume that we talking about the relevant
`.gitignore` file (where the pattern appears in).

Perhaps this covers it all?

        A pattern with a non-trailing slash is always considered
        to begin at the `.gitignore` file location.

followed by your example

> For example, the pattern `doc/frotz/` that appears in
> `.gitignore` at the top-level of the project matches
> `doc/frotz` directory (again, seen from the top-level), but
> not `a/doc/frotz`.

and maybe one more example

        Note that the pattern `doc/frotz` and `/doc/frotz` are
        equivalent.
        However `/bar` and `bar` are different. They both match the
        `bar` file or folder at the top level, but only the latter
will also match `foo/bar`
        (when `foo` is at the top level).

This avoids the hustle with the ambiguous path, where it starts, and
trailing or leading slashes. Together with the
two examples it seems to be a good compromise between accuracy and
understandable.

The alternative would be to say

        A pattern with a non-trailing slash is only matched against any
        path that begins in the directory of the `.gitignore` file.

While this is maybe clearer then saying "pattern [...] always
considered to begin at` it is ambiguous about the slashes.
So a very accuracy but maybe less understandable version would be
something like this:

        A pattern with a non-trailing slash is only matched against any
        path that begins in the directory of the `.gitignore` file.
        For example, if the `.gitignore` file is in folder `doc`
        the path to file  `bar/doc/a/foo` that begins in `doc` is `a/foo`.
        A pattern that matches a path except for a leading slash or
trailing slash
        is still considered a match. It is still valid however,
        that when a pattern ends with a slash, it would only find a
match with a directory.

---------

> Also, a pattern "/doc" matches doc at the current level (i.e. the
> directory in which .gitignore file that the pattern was taken from
> is found) and not in any subdirectories.  Is that clear in the
> proposed update?

Yes.

However, in the docs is already one paragraph solely dedicated for this case:

> A leading slash matches the beginning of the pathname. For example, "/*.c" matches
> "cat-file.c" but not "mozilla-sha1/sha1.c".

However, we have already a better and more in detail explained example
in the new
proposal `*` paragraph and and the case with the leading slash is
now a sub-case of `A pattern with a non-trailing slash`
so we might just get rid of the above paragraph?
----------


Thank you for explaining me how the algorithm works procedurally.
It gave some inside of the origin of "If the pattern ends with a slash, it is
removed for the purpose of the following description.."
---------

All the best,
Adam

Re: [PATCH/docs] make slash-rules more readable

[Dr. Adam Nielsen]

I think its maybe hard to track all the changes that we have discussed
so far. Should I create a new PATCH request including all the changes
from the recent mails and then we continue the discussion from there?

Best regards,
Adam

Am Mi., 10. Apr. 2019 um 09:39 Uhr schrieb Dr. Adam Nielsen <admin@in-ici.net>:
>
> > the pattern is matched against paths in the directory where the
> > `.gitignore` file that has the pattern in it is in, and any of
> > its subdirectories (recursively).
>
> > the pattern will match in all directories relative to
> > the `.gitignore` file, with infinite depth.
>
> I could not catch the difference between the meaning of both.
> However, I think "paths in the directory" and "directories relative to"
> are maybe both ambiguous.
>
> Since a pattern without a non-trailing slash must always be a file name or a
> folder name, and does not have a leading slash, we could maybe just
> say it like this:
>
>         the pattern is matched against all files and folders (recursively)
>         from the location of the `.gitignore` file.
> ---------
>
>
> > Unlike a pattern without a slash, a pattern with a
> > non-trailing slash is matched against paths immediately in
> > the directory the `.gitignore` file the pattern appears in
> > is stored in, and does not get used in its subdirectories..
>
> I think one can always assume that we talking about the relevant
> `.gitignore` file (where the pattern appears in).
>
> Perhaps this covers it all?
>
>         A pattern with a non-trailing slash is always considered
>         to begin at the `.gitignore` file location.
>
> followed by your example
>
> > For example, the pattern `doc/frotz/` that appears in
> > `.gitignore` at the top-level of the project matches
> > `doc/frotz` directory (again, seen from the top-level), but
> > not `a/doc/frotz`.
>
> and maybe one more example
>
>         Note that the pattern `doc/frotz` and `/doc/frotz` are
>         equivalent.
>         However `/bar` and `bar` are different. They both match the
>         `bar` file or folder at the top level, but only the latter
> will also match `foo/bar`
>         (when `foo` is at the top level).
>
> This avoids the hustle with the ambiguous path, where it starts, and
> trailing or leading slashes. Together with the
> two examples it seems to be a good compromise between accuracy and
> understandable.
>
> The alternative would be to say
>
>         A pattern with a non-trailing slash is only matched against any
>         path that begins in the directory of the `.gitignore` file.
>
> While this is maybe clearer then saying "pattern [...] always
> considered to begin at` it is ambiguous about the slashes.
> So a very accuracy but maybe less understandable version would be
> something like this:
>
>         A pattern with a non-trailing slash is only matched against any
>         path that begins in the directory of the `.gitignore` file.
>         For example, if the `.gitignore` file is in folder `doc`
>         the path to file  `bar/doc/a/foo` that begins in `doc` is `a/foo`.
>         A pattern that matches a path except for a leading slash or
> trailing slash
>         is still considered a match. It is still valid however,
>         that when a pattern ends with a slash, it would only find a
> match with a directory.
>
> ---------
>
> > Also, a pattern "/doc" matches doc at the current level (i.e. the
> > directory in which .gitignore file that the pattern was taken from
> > is found) and not in any subdirectories.  Is that clear in the
> > proposed update?
>
> Yes.
>
> However, in the docs is already one paragraph solely dedicated for this case:
>
> > A leading slash matches the beginning of the pathname. For example, "/*.c" matches
> > "cat-file.c" but not "mozilla-sha1/sha1.c".
>
> However, we have already a better and more in detail explained example
> in the new
> proposal `*` paragraph and and the case with the leading slash is
> now a sub-case of `A pattern with a non-trailing slash`
> so we might just get rid of the above paragraph?
> ----------
>
>
> Thank you for explaining me how the algorithm works procedurally.
> It gave some inside of the origin of "If the pattern ends with a slash, it is
> removed for the purpose of the following description.."
> ---------
>
> All the best,
> Adam