Re: [PATCH v2 1/2] abspath: add a function to resolve paths with missing components

[Johannes Schindelin <Johannes.Schindelin@gmx.de>]

Hi brian,

On Fri, 9 Oct 2020, brian m. carlson wrote:

> We'd like to canonicalize paths such that we can preserve any number of
> trailing components that may be missing.  Let's add a function to do
> that that calls strbuf_realpath to find the canonical path for the
> portion we do have and then append the missing part.  We adjust
> strip_last_component to return us the component it has stripped and use
> that to help us accumulate the missing part.
>
> Note that it is intentional that we invoke strbuf_realpath here,
> repeatedly if necessary, because on Windows that function is replaced
> with a version that uses the proper system semantics for
> canonicalization.  Trying to adjust strbuf_realpath to perform this kind
> of canonicalization with an additional option would fail to work
> properly on Windows.  The present approach is equivalent to
> strbuf_realpath for cases where the path exists, and the only other
> cases where we will use this function the additional overhead of
> multiple invocations is not significant.

Thank you for being so considerate. Yes, on Windows, we use (wherever
possible) a shortcut that tells us the canonicalized path of existing
entries.

Technically, it is not `strbuf_realpath()` that we override, but we take a
shortcut _in_ that function. That's semantics, though.

More importantly, we recently fixed a bug in our code to allow for a quirk
in the `strbuf_realpath()` function: `strbuf_realpath()` allows the last
path component to not exist. If that is the case, now it's time to try
without last component.

In a sense, this is a 1-level version of your infinite-level
`strbuf_realpath_missing()` function.

An idea that immediately crosses my mind is whether that level could be
something we want to pass directly into `strbuf_realpath()` as a parameter
(it would be 1 to imitate the current behavior and -1 for the
infinite-level case). What do you think? Does that make sense?

In any case, I think this `_missing()` functionality should be implemented
a bit more tightly with the `strbuf_realpath()` function because of the
logic that already allows the last component to be missing:

                if (lstat(resolved->buf, &st)) {
                        /* error out unless this was the last component */
                        if (errno != ENOENT || remaining.len) {
                                if (die_on_error)
                                        die_errno("Invalid path '%s'",
                                                  resolved->buf);
                                else
                                        goto error_out;
                        }

See https://github.com/git/git/blob/v2.29.2/abspath.c#L130-L138 for the
exact code and context.

Seeing as we _already_ have some code to allow for _some_ missing
component, it should be possible to extend the logic to allow for
different levels (e.g. using `count_slashes()` if we want to allow more
than just the last component to be missing).

> Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
> ---
>  abspath.c | 50 ++++++++++++++++++++++++++++++++++++++++++++++----
>  cache.h   |  1 +
>  2 files changed, 47 insertions(+), 4 deletions(-)
>
> diff --git a/abspath.c b/abspath.c
> index 6f15a418bb..092bb33b64 100644
> --- a/abspath.c
> +++ b/abspath.c
> @@ -11,8 +11,12 @@ int is_directory(const char *path)
>  	return (!stat(path, &st) && S_ISDIR(st.st_mode));
>  }
>
> -/* removes the last path component from 'path' except if 'path' is root */
> -static void strip_last_component(struct strbuf *path)
> +/*
> + * Removes the last path component from 'path' except if 'path' is root.
> + *
> + * If last is not NULL, the last path component is copied to last.
> + */
> +static void strip_last_component(struct strbuf *path, struct strbuf *last)
>  {
>  	size_t offset = offset_1st_component(path->buf);
>  	size_t len = path->len;
> @@ -20,6 +24,10 @@ static void strip_last_component(struct strbuf *path)
>  	/* Find start of the last component */
>  	while (offset < len && !is_dir_sep(path->buf[len - 1]))
>  		len--;
> +
> +	if (last)
> +		strbuf_addstr(last, path->buf + len);
> +
>  	/* Skip sequences of multiple path-separators */
>  	while (offset < len && is_dir_sep(path->buf[len - 1]))
>  		len--;
> @@ -118,7 +126,7 @@ char *strbuf_realpath(struct strbuf *resolved, const char *path,
>  			continue; /* '.' component */
>  		} else if (next.len == 2 && !strcmp(next.buf, "..")) {
>  			/* '..' component; strip the last path component */
> -			strip_last_component(resolved);
> +			strip_last_component(resolved, NULL);
>  			continue;
>  		}
>
> @@ -169,7 +177,7 @@ char *strbuf_realpath(struct strbuf *resolved, const char *path,
>  				 * strip off the last component since it will
>  				 * be replaced with the contents of the symlink
>  				 */
> -				strip_last_component(resolved);
> +				strip_last_component(resolved, NULL);
>  			}
>
>  			/*
> @@ -202,6 +210,40 @@ char *strbuf_realpath(struct strbuf *resolved, const char *path,
>  	return retval;
>  }
>
> +/*
> + * Like strbuf_realpath, but trailing components which do not exist are copied
> + * through.
> + */
> +char *strbuf_realpath_missing(struct strbuf *resolved, const char *path)
> +{
> +	struct strbuf remaining = STRBUF_INIT;
> +	struct strbuf trailing = STRBUF_INIT;
> +	struct strbuf component = STRBUF_INIT;
> +
> +	strbuf_addstr(&remaining, path);
> +
> +	while (remaining.len) {
> +		if (strbuf_realpath(resolved, remaining.buf, 0)) {
> +			strbuf_addbuf(resolved, &trailing);
> +
> +			strbuf_release(&component);
> +			strbuf_release(&remaining);
> +			strbuf_release(&trailing);
> +
> +			return resolved->buf;
> +		}
> +		strip_last_component(&remaining, &component);
> +		strbuf_insertstr(&trailing, 0, "/");
> +		strbuf_insertstr(&trailing, 1, component.buf);

Not that it matters a lot, but this could be written shorter via

		strbuf_insertf(&trailing, "/%s", component.buf);

But as I said above, I think we should be able to fold the logic _into_
`strbuf_realpath()` (even if this makes my job harder to maintain the
Windows-specific shortcut).

Thanks,
Dscho

> +		strbuf_reset(&component);
> +	}
> +
> +	strbuf_release(&component);
> +	strbuf_release(&remaining);
> +	strbuf_release(&trailing);
> +	return NULL;
> +}
> +
>  char *real_pathdup(const char *path, int die_on_error)
>  {
>  	struct strbuf realpath = STRBUF_INIT;
> diff --git a/cache.h b/cache.h
> index c0072d43b1..e1e17e108e 100644
> --- a/cache.h
> +++ b/cache.h
> @@ -1320,6 +1320,7 @@ static inline int is_absolute_path(const char *path)
>  int is_directory(const char *);
>  char *strbuf_realpath(struct strbuf *resolved, const char *path,
>  		      int die_on_error);
> +char *strbuf_realpath_missing(struct strbuf *resolved, const char *path);
>  char *real_pathdup(const char *path, int die_on_error);
>  const char *absolute_path(const char *path);
>  char *absolute_pathdup(const char *path);
>