diff mbox series

[RFC,v2,1/1] refspec: add support for negative refspecs

Message ID 20200821215247.758978-2-jacob.e.keller@intel.com
State New, archived
Headers show
Series implement support for negative refspecs | expand

Commit Message

Jacob Keller Aug. 21, 2020, 9:52 p.m. UTC
From: Jacob Keller <jacob.keller@gmail.com>

Both fetch and push support pattern refspecs which allow fetching or
pushing references that match a specific pattern. Because these patterns
are globs, they have somewhat limited ability to express more complex
situations.

For example, suppose you wish to fetch all branches from a remote except
for a specific one. To allow this, you must setup a set of refspecs
which match only the branches you want. Because refspecs are either
explicit name matches, or simple globs, many patterns cannot be
expressed.

Add support for a new type of refspec, referred to as "negative"
refspecs. These are prefixed with a '^' and mean "exclude any ref
matching this refspec". They can only have one "side" which always
refers to the source. During a fetch, this refers to the name of the ref
on the remote. During a push, this refers to the name of the ref on the
local side.

With negative refspecs, users can express more complex patterns. For
example:

 git fetch origin refs/heads/*:refs/remotes/origin/* ^refs/heads/dontwant

will fetch all branches on origin into remotes/origin, but will exclude
fetching the branch named dontwant.

Refspecs today are commutative, meaning that order doesn't expressly
matter. Rather than forcing an implied order, negative refspecs will
always be applied last. That is, in order to match, a ref must match at
least one positive refspec, and match none of the negative refspecs.
This is similar to how negative pathspecs work.

Signed-off-by: Jacob Keller <jacob.keller@gmail.com>
---

I realize this probably needs to be broken down into multiple patches, but I
haven't quite figured out the best way to do that. I'd like to avoid the
case where a commit has support for parsing negative refspecs but code paths
which use refspecs aren't handling them correctly. Thoughts?

Splitting would also allow additional space for explanations of some of the
trickier logic.

I am also definitely looking for more test ideas, to help make sure we
cover a good variety of the flows.

 builtin/fetch.c                   |  10 ++
 refspec.c                         |  34 ++++++-
 refspec.h                         |  14 +--
 remote.c                          | 139 +++++++++++++++++++++++++--
 remote.h                          |   9 +-
 t/t5582-fetch-negative-refspec.sh | 151 ++++++++++++++++++++++++++++++
 6 files changed, 340 insertions(+), 17 deletions(-)
 create mode 100755 t/t5582-fetch-negative-refspec.sh

Comments

Johannes Schindelin Aug. 22, 2020, 1:29 p.m. UTC | #1
Hi Jake,

On Fri, 21 Aug 2020, Jacob Keller wrote:

>  static void query_refspecs_multiple(struct refspec *rs,
>  				    struct refspec_item *query,
>  				    struct string_list *results)
>  {
> -	int i;
> +	int i, matched_negative = 0;
>  	int find_src = !query->src;
> +	struct string_list reversed = STRING_LIST_INIT_NODUP;
> +	const char *needle = find_src ? query->dst : query->src;
> +	char **result = find_src ? &query->src : &query->dst;
>
>  	if (find_src && !query->dst)
>  		BUG("query_refspecs_multiple: need either src or dst");
>
> +	/*
> +	 * If a ref matches any of the negative refspecs, then we should treat
> +	 * it as not matching this query. Note that negative refspecs apply to
> +	 * the source but we're checking only the destination. Reverse and
> +	 * capture any pattern refspecs in order to see if the source would
> +	 * have matched a negative refspec.
> +	 */
> +	for (i = 0; i < rs->nr; i++) {
> +		struct refspec_item *refspec = &rs->items[i];
> +		char *expn_name;
> +
> +		if (refspec->negative)
> +			continue;
> +
> +		/* Note the reversal of src and dst */
> +		if (refspec->pattern) {
> +			const char *key = refspec->dst ?: refspec->src;

This breaks at least with Visual C. I need this to unbreak the build:

-- snipsnap --
Subject: [PATCH] fixup! refspec: add support for negative refspecs

The `?:` operator is not supported e.g. by Visual C. Let's not use it.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
---
 remote.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/remote.c b/remote.c
index 940622e0ba88..8e081863c06d 100644
--- a/remote.c
+++ b/remote.c
@@ -731,7 +731,7 @@ static void query_refspecs_multiple(struct refspec *rs,

 		/* Note the reversal of src and dst */
 		if (refspec->pattern) {
-			const char *key = refspec->dst ?: refspec->src;
+			const char *key = refspec->dst ? refspec->dst : refspec->src;
 			const char *value = refspec->src;

 			if (match_name_with_pattern(key, needle, value, &expn_name))
@@ -795,7 +795,7 @@ int query_refspecs(struct refspec *rs, struct refspec_item *query)

 		/* Note the reversal of src and dst */
 		if (refspec->pattern) {
-			const char *key = refspec->dst ?: refspec->src;
+			const char *key = refspec->dst ? refspec->dst : refspec->src;
 			const char *value = refspec->src;

 			if (match_name_with_pattern(key, needle, value, &expn_name))
--
2.28.0.windows.1
Jacob Keller Aug. 24, 2020, 3:47 p.m. UTC | #2
On Sun, Aug 23, 2020 at 1:29 PM Johannes Schindelin
<Johannes.Schindelin@gmx.de> wrote:
>
> Hi Jake,
>
> On Fri, 21 Aug 2020, Jacob Keller wrote:
>
> >  static void query_refspecs_multiple(struct refspec *rs,
> >                                   struct refspec_item *query,
> >                                   struct string_list *results)
> >  {
> > -     int i;
> > +     int i, matched_negative = 0;
> >       int find_src = !query->src;
> > +     struct string_list reversed = STRING_LIST_INIT_NODUP;
> > +     const char *needle = find_src ? query->dst : query->src;
> > +     char **result = find_src ? &query->src : &query->dst;
> >
> >       if (find_src && !query->dst)
> >               BUG("query_refspecs_multiple: need either src or dst");
> >
> > +     /*
> > +      * If a ref matches any of the negative refspecs, then we should treat
> > +      * it as not matching this query. Note that negative refspecs apply to
> > +      * the source but we're checking only the destination. Reverse and
> > +      * capture any pattern refspecs in order to see if the source would
> > +      * have matched a negative refspec.
> > +      */
> > +     for (i = 0; i < rs->nr; i++) {
> > +             struct refspec_item *refspec = &rs->items[i];
> > +             char *expn_name;
> > +
> > +             if (refspec->negative)
> > +                     continue;
> > +
> > +             /* Note the reversal of src and dst */
> > +             if (refspec->pattern) {
> > +                     const char *key = refspec->dst ?: refspec->src;
>
> This breaks at least with Visual C. I need this to unbreak the build:
>


The provided fixup! looks good to me. I'll pull it in when making the
next revision.

Thanks,
Jake

> -- snipsnap --
> Subject: [PATCH] fixup! refspec: add support for negative refspecs
>
> The `?:` operator is not supported e.g. by Visual C. Let's not use it.
>
> Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
> ---
>  remote.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/remote.c b/remote.c
> index 940622e0ba88..8e081863c06d 100644
> --- a/remote.c
> +++ b/remote.c
> @@ -731,7 +731,7 @@ static void query_refspecs_multiple(struct refspec *rs,
>
>                 /* Note the reversal of src and dst */
>                 if (refspec->pattern) {
> -                       const char *key = refspec->dst ?: refspec->src;
> +                       const char *key = refspec->dst ? refspec->dst : refspec->src;
>                         const char *value = refspec->src;
>
>                         if (match_name_with_pattern(key, needle, value, &expn_name))
> @@ -795,7 +795,7 @@ int query_refspecs(struct refspec *rs, struct refspec_item *query)
>
>                 /* Note the reversal of src and dst */
>                 if (refspec->pattern) {
> -                       const char *key = refspec->dst ?: refspec->src;
> +                       const char *key = refspec->dst ? refspec->dst : refspec->src;
>                         const char *value = refspec->src;
>
>                         if (match_name_with_pattern(key, needle, value, &expn_name))
> --
> 2.28.0.windows.1
>
Junio C Hamano Aug. 24, 2020, 5:55 p.m. UTC | #3
Johannes Schindelin <Johannes.Schindelin@gmx.de> writes:

> The `?:` operator is not supported e.g. by Visual C. Let's not use it.

Thanks.  IIRC that was GNU extension?
Jacob Keller Aug. 24, 2020, 7:26 p.m. UTC | #4
On Mon, Aug 24, 2020 at 10:55 AM Junio C Hamano <gitster@pobox.com> wrote:
>
> Johannes Schindelin <Johannes.Schindelin@gmx.de> writes:
>
> > The `?:` operator is not supported e.g. by Visual C. Let's not use it.
>
> Thanks.  IIRC that was GNU extension?
>

Yea, probably.

From wikipedia, [1], "In GNU C and C++ (that is: in C and C++ with GCC
extensions), the second operand of the ternary operator is optional.
This has been the case since at least GCC 2.95.3 (March 2001), and
seems to the original elvis operator."

Thanks,
Jake

[1] https://en.wikipedia.org/wiki/Elvis_operator
Junio C Hamano Sept. 17, 2020, 8:21 p.m. UTC | #5
Jacob Keller <jacob.e.keller@intel.com> writes:

> From: Jacob Keller <jacob.keller@gmail.com>
>
> Both fetch and push support pattern refspecs which allow fetching or
> pushing references that match a specific pattern. Because these patterns
> are globs, they have somewhat limited ability to express more complex
> situations.
>
> For example, suppose you wish to fetch all branches from a remote except
> for a specific one. To allow this, you must setup a set of refspecs
> which match only the branches you want. Because refspecs are either
> explicit name matches, or simple globs, many patterns cannot be
> expressed.
>
> Add support for a new type of refspec, referred to as "negative"
> refspecs. These are prefixed with a '^' and mean "exclude any ref
> matching this refspec". They can only have one "side" which always
> refers to the source. During a fetch, this refers to the name of the ref
> on the remote. During a push, this refers to the name of the ref on the
> local side.
>
> With negative refspecs, users can express more complex patterns. For
> example:
>
>  git fetch origin refs/heads/*:refs/remotes/origin/* ^refs/heads/dontwant
>
> will fetch all branches on origin into remotes/origin, but will exclude
> fetching the branch named dontwant.
>
> Refspecs today are commutative, meaning that order doesn't expressly
> matter. Rather than forcing an implied order, negative refspecs will
> always be applied last. That is, in order to match, a ref must match at
> least one positive refspec, and match none of the negative refspecs.
> This is similar to how negative pathspecs work.
>
> Signed-off-by: Jacob Keller <jacob.keller@gmail.com>
> ---
>
> I realize this probably needs to be broken down into multiple patches, but I
> haven't quite figured out the best way to do that. I'd like to avoid the
> case where a commit has support for parsing negative refspecs but code paths
> which use refspecs aren't handling them correctly. Thoughts?
>
> Splitting would also allow additional space for explanations of some of the
> trickier logic.
>
> I am also definitely looking for more test ideas, to help make sure we
> cover a good variety of the flows.

Anybody wants to help this move forward?

I plan to send a review with the patch in the current form, without
waiting for any splitting, later towards the weekend, though.
Junio C Hamano Sept. 18, 2020, 12:01 a.m. UTC | #6
Jacob Keller <jacob.e.keller@intel.com> writes:

> @@ -66,6 +74,28 @@ static int parse_refspec(struct refspec_item *item, const char *refspec, int fet
>  	item->src = xstrndup(lhs, llen);
>  	flags = REFNAME_ALLOW_ONELEVEL | (is_glob ? REFNAME_REFSPEC_PATTERN : 0);
>  
> +	if (item->negative) {
> +		struct object_id unused;
> +
> +		/*
> +		 * Negative refspecs only have a LHS, which indicates a ref
> +		 * (or pattern of refs) to exclude from other matches. This
> +		 * can either be a simple ref, a glob pattern, or even an
> +		 * exact sha1 match.
> +		 */
> +		if (!*item->src)
> +			return 0; /* negative refspecs must not be empty */
> +		else if (llen == the_hash_algo->hexsz && !get_oid_hex(item->src, &unused))
> +			item->exact_sha1 = 1; /* ok */
> +		else if (!check_refname_format(item->src, flags))
> +			; /* valid looking ref is ok */
> +		else
> +			return 0;
> +
> +		/* other rules for negative refspecs don't apply */

This comment confused me a bit; did you mean "other rules don't
apply to negative refspecs"?

> +		return 1;
> +	}
> +
>  	if (fetch) {
>  		struct object_id unused;


> diff --git a/remote.c b/remote.c
> index c5ed74f91c63..2f583d72c3f0 100644
> --- a/remote.c
> +++ b/remote.c
> @@ -1058,7 +1172,7 @@ static int match_explicit(struct ref *src, struct ref *dst,
>  	const char *dst_value = rs->dst;
>  	char *dst_guess;
>  
> -	if (rs->pattern || rs->matching)
> +	if (rs->pattern || rs->matching || rs->negative)
>  		return 0;

OK.  These "special" ones do not participate in explicit matching.

> @@ -1134,6 +1248,10 @@ static char *get_ref_match(const struct refspec *rs, const struct ref *ref,
>  	int matching_refs = -1;
>  	for (i = 0; i < rs->nr; i++) {
>  		const struct refspec_item *item = &rs->items[i];
> +
> +		if (item->negative)
> +			continue;
> +

And a negative one does not decide if a ref being pushed will be
pushed out for real at this point.  This helper is only to enumerate
the candidate refs to be pushed out; the caller makes a separate
call to apply_negative_refspecs() to cull the candidate list later.

OK.

> @@ -1339,7 +1457,7 @@ int check_push_refs(struct ref *src, struct refspec *rs)
>  	for (i = 0; i < rs->nr; i++) {
>  		struct refspec_item *item = &rs->items[i];
>  
> -		if (item->pattern || item->matching)
> +		if (item->pattern || item->matching || item->negative)
>  			continue;
>  
>  		ret |= match_explicit_lhs(src, item, NULL, NULL);

match_explicit_lhs(), like match_explicit(), are for explicit
matching and should not be called for the "special" ones.  OK.

> @@ -1441,6 +1559,8 @@ int match_push_refs(struct ref *src, struct ref **dst,
>  		string_list_clear(&src_ref_index, 0);
>  	}
>  
> +	*dst = apply_negative_refspecs(*dst, rs);
> +
>  	if (errs)
>  		return -1;
>  	return 0;

And after grabbing all the candidate refs to be updated via this
push, we filter out the ones that match negative pattern.  Can it
also produce an error, or it can never fail (to udpate errs)?

> @@ -1810,6 +1930,9 @@ int get_fetch_map(const struct ref *remote_refs,
>  {
>  	struct ref *ref_map, **rmp;
>  
> +	if (refspec->negative)
> +		return 0;
> +

Again, the idea is to let the existing codepath to only deal with
the positive refspec elements to keep the same behaviour, and let
the caller filter the ones that match negative ones out of the
result.  So we return without anything here for negative one.

Nothing jumped out at me as being suspicious so far, other than that
the GNU "?<empty>:" thing needs to be fixed as pointed out by Dscho.

Thanks.
Jacob Keller Sept. 24, 2020, 11:33 p.m. UTC | #7
On Thu, Sep 17, 2020 at 5:02 PM Junio C Hamano <gitster@pobox.com> wrote:
>
> Jacob Keller <jacob.e.keller@intel.com> writes:
>
> > @@ -66,6 +74,28 @@ static int parse_refspec(struct refspec_item *item, const char *refspec, int fet
> >       item->src = xstrndup(lhs, llen);
> >       flags = REFNAME_ALLOW_ONELEVEL | (is_glob ? REFNAME_REFSPEC_PATTERN : 0);
> >
> > +     if (item->negative) {
> > +             struct object_id unused;
> > +
> > +             /*
> > +              * Negative refspecs only have a LHS, which indicates a ref
> > +              * (or pattern of refs) to exclude from other matches. This
> > +              * can either be a simple ref, a glob pattern, or even an
> > +              * exact sha1 match.
> > +              */
> > +             if (!*item->src)
> > +                     return 0; /* negative refspecs must not be empty */
> > +             else if (llen == the_hash_algo->hexsz && !get_oid_hex(item->src, &unused))
> > +                     item->exact_sha1 = 1; /* ok */
> > +             else if (!check_refname_format(item->src, flags))
> > +                     ; /* valid looking ref is ok */
> > +             else
> > +                     return 0;
> > +
> > +             /* other rules for negative refspecs don't apply */
>
> This comment confused me a bit; did you mean "other rules don't
> apply to negative refspecs"?
>

Yea, this should be reworded.

> > +             return 1;
> > +     }
> > +
> >       if (fetch) {
> >               struct object_id unused;
>
>
> > diff --git a/remote.c b/remote.c
> > index c5ed74f91c63..2f583d72c3f0 100644
> > --- a/remote.c
> > +++ b/remote.c
> > @@ -1058,7 +1172,7 @@ static int match_explicit(struct ref *src, struct ref *dst,
> >       const char *dst_value = rs->dst;
> >       char *dst_guess;
> >
> > -     if (rs->pattern || rs->matching)
> > +     if (rs->pattern || rs->matching || rs->negative)
> >               return 0;
>
> OK.  These "special" ones do not participate in explicit matching.
>
> > @@ -1134,6 +1248,10 @@ static char *get_ref_match(const struct refspec *rs, const struct ref *ref,
> >       int matching_refs = -1;
> >       for (i = 0; i < rs->nr; i++) {
> >               const struct refspec_item *item = &rs->items[i];
> > +
> > +             if (item->negative)
> > +                     continue;
> > +
>
> And a negative one does not decide if a ref being pushed will be
> pushed out for real at this point.  This helper is only to enumerate
> the candidate refs to be pushed out; the caller makes a separate
> call to apply_negative_refspecs() to cull the candidate list later.
>
> OK.
>
> > @@ -1339,7 +1457,7 @@ int check_push_refs(struct ref *src, struct refspec *rs)
> >       for (i = 0; i < rs->nr; i++) {
> >               struct refspec_item *item = &rs->items[i];
> >
> > -             if (item->pattern || item->matching)
> > +             if (item->pattern || item->matching || item->negative)
> >                       continue;
> >
> >               ret |= match_explicit_lhs(src, item, NULL, NULL);
>
> match_explicit_lhs(), like match_explicit(), are for explicit
> matching and should not be called for the "special" ones.  OK.
>
> > @@ -1441,6 +1559,8 @@ int match_push_refs(struct ref *src, struct ref **dst,
> >               string_list_clear(&src_ref_index, 0);
> >       }
> >
> > +     *dst = apply_negative_refspecs(*dst, rs);
> > +
> >       if (errs)
> >               return -1;
> >       return 0;
>
> And after grabbing all the candidate refs to be updated via this
> push, we filter out the ones that match negative pattern.  Can it
> also produce an error, or it can never fail (to udpate errs)?
>
> > @@ -1810,6 +1930,9 @@ int get_fetch_map(const struct ref *remote_refs,
> >  {
> >       struct ref *ref_map, **rmp;
> >
> > +     if (refspec->negative)
> > +             return 0;
> > +
>
> Again, the idea is to let the existing codepath to only deal with
> the positive refspec elements to keep the same behaviour, and let
> the caller filter the ones that match negative ones out of the
> result.  So we return without anything here for negative one.
>

Yep, that's what I went for. The only real downside here is if we
forget a code path that should honor negative refspecs and doesn't,
because it will "accept" the refspec list with such a negative
refspec, but not do anything with it.

> Nothing jumped out at me as being suspicious so far, other than that
> the GNU "?<empty>:" thing needs to be fixed as pointed out by Dscho.
>
> Thanks.
Jacob Keller Sept. 24, 2020, 11:42 p.m. UTC | #8
On Thu, Sep 17, 2020 at 5:02 PM Junio C Hamano <gitster@pobox.com> wrote:
>
> Jacob Keller <jacob.e.keller@intel.com> writes:
>
> > @@ -1441,6 +1559,8 @@ int match_push_refs(struct ref *src, struct ref **dst,
> >               string_list_clear(&src_ref_index, 0);
> >       }
> >
> > +     *dst = apply_negative_refspecs(*dst, rs);
> > +
> >       if (errs)
> >               return -1;
> >       return 0;
>
> And after grabbing all the candidate refs to be updated via this
> push, we filter out the ones that match negative pattern.  Can it
> also produce an error, or it can never fail (to udpate errs)?
>

It can't fail. Either refs match a negative refspec and will get
excluded, or they do not match a negative refspec and will be kept. We
already validate the negative refspecs earlier.

Thanks,
Jake
diff mbox series

Patch

diff --git a/builtin/fetch.c b/builtin/fetch.c
index c49f0e975203..310699eb7e45 100644
--- a/builtin/fetch.c
+++ b/builtin/fetch.c
@@ -530,6 +530,16 @@  static struct ref *get_ref_map(struct remote *remote,
 		tail = &rm->next;
 	}
 
+	/*
+	 * apply negative refspecs first, before we remove duplicates. This is
+	 * necessary as negative refspecs might remove an otherwise conflicting
+	 * duplicate.
+	 */
+	if (rs->nr)
+		ref_map = apply_negative_refspecs(ref_map, rs);
+	else
+		ref_map = apply_negative_refspecs(ref_map, &remote->fetch);
+
 	ref_map = ref_remove_duplicates(ref_map);
 
 	refname_hash_init(&existing_refs);
diff --git a/refspec.c b/refspec.c
index f10ef284cef9..25755b83c5f2 100644
--- a/refspec.c
+++ b/refspec.c
@@ -8,6 +8,7 @@  static struct refspec_item s_tag_refspec = {
 	1,
 	0,
 	0,
+	0,
 	"refs/tags/*",
 	"refs/tags/*"
 };
@@ -32,10 +33,17 @@  static int parse_refspec(struct refspec_item *item, const char *refspec, int fet
 	if (*lhs == '+') {
 		item->force = 1;
 		lhs++;
+	} else if (*lhs == '^') {
+		item->negative = 1;
+		lhs++;
 	}
 
 	rhs = strrchr(lhs, ':');
 
+	/* negative refspecs only have one side */
+	if (item->negative && rhs)
+		return 0;
+
 	/*
 	 * Before going on, special case ":" (or "+:") as a refspec
 	 * for pushing matching refs.
@@ -55,7 +63,7 @@  static int parse_refspec(struct refspec_item *item, const char *refspec, int fet
 
 	llen = (rhs ? (rhs - lhs - 1) : strlen(lhs));
 	if (1 <= llen && memchr(lhs, '*', llen)) {
-		if ((rhs && !is_glob) || (!rhs && fetch))
+		if ((rhs && !is_glob) || (!rhs && !item->negative && fetch))
 			return 0;
 		is_glob = 1;
 	} else if (rhs && is_glob) {
@@ -66,6 +74,28 @@  static int parse_refspec(struct refspec_item *item, const char *refspec, int fet
 	item->src = xstrndup(lhs, llen);
 	flags = REFNAME_ALLOW_ONELEVEL | (is_glob ? REFNAME_REFSPEC_PATTERN : 0);
 
+	if (item->negative) {
+		struct object_id unused;
+
+		/*
+		 * Negative refspecs only have a LHS, which indicates a ref
+		 * (or pattern of refs) to exclude from other matches. This
+		 * can either be a simple ref, a glob pattern, or even an
+		 * exact sha1 match.
+		 */
+		if (!*item->src)
+			return 0; /* negative refspecs must not be empty */
+		else if (llen == the_hash_algo->hexsz && !get_oid_hex(item->src, &unused))
+			item->exact_sha1 = 1; /* ok */
+		else if (!check_refname_format(item->src, flags))
+			; /* valid looking ref is ok */
+		else
+			return 0;
+
+		/* other rules for negative refspecs don't apply */
+		return 1;
+	}
+
 	if (fetch) {
 		struct object_id unused;
 
@@ -209,7 +239,7 @@  void refspec_ref_prefixes(const struct refspec *rs,
 		const struct refspec_item *item = &rs->items[i];
 		const char *prefix = NULL;
 
-		if (item->exact_sha1)
+		if (item->exact_sha1 || item->negative)
 			continue;
 		if (rs->fetch == REFSPEC_FETCH)
 			prefix = item->src;
diff --git a/refspec.h b/refspec.h
index 8d654e3a3ac4..e5bf6d25d0f7 100644
--- a/refspec.h
+++ b/refspec.h
@@ -5,12 +5,13 @@ 
 extern const struct refspec_item *tag_refspec;
 
 /**
- * A struct refspec_item holds the parsed interpretation of a refspec.  If it will
- * force updates (starts with a '+'), force is true.  If it is a pattern
- * (sides end with '*') pattern is true.  src and dest are the two sides
- * (including '*' characters if present); if there is only one side, it is src,
- * and dst is NULL; if sides exist but are empty (i.e., the refspec either
- * starts or ends with ':'), the corresponding side is "".
+ * A struct refspec_item holds the parsed interpretation of a refspec.  If it
+ * will force updates (starts with a '+'), force is true.  If it is a pattern
+ * (sides end with '*') pattern is true.  If it is a negative refspec, (starts
+ * with '^'), negative is true.  src and dest are the two sides (including '*'
+ * characters if present); if there is only one side, it is src, and dst is
+ * NULL; if sides exist but are empty (i.e., the refspec either starts or ends
+ * with ':'), the corresponding side is "".
  *
  * remote_find_tracking(), given a remote and a struct refspec_item with either src
  * or dst filled out, will fill out the other such that the result is in the
@@ -22,6 +23,7 @@  struct refspec_item {
 	unsigned pattern : 1;
 	unsigned matching : 1;
 	unsigned exact_sha1 : 1;
+	unsigned negative : 1;
 
 	char *src;
 	char *dst;
diff --git a/remote.c b/remote.c
index c5ed74f91c63..2f583d72c3f0 100644
--- a/remote.c
+++ b/remote.c
@@ -686,24 +686,100 @@  static int match_name_with_pattern(const char *key, const char *name,
 	return ret;
 }
 
+static int refspec_match(const struct refspec_item *refspec,
+			 const char *name)
+{
+	if (refspec->pattern)
+		return match_name_with_pattern(refspec->src, name, NULL, NULL);
+
+	return !strcmp(refspec->src, name);
+}
+
+static int omit_name_by_refspec(const char *name, struct refspec *rs)
+{
+	int i;
+
+	for (i = 0; i < rs->nr; i++) {
+		if (rs->items[i].negative && refspec_match(&rs->items[i], name))
+			return 1;
+	}
+	return 0;
+}
+
+struct ref *apply_negative_refspecs(struct ref *ref_map, struct refspec *rs)
+{
+	struct ref **tail;
+
+	for (tail = &ref_map; *tail; ) {
+		struct ref *ref = *tail;
+
+		if (omit_name_by_refspec(ref->name, rs)) {
+			*tail = ref->next;
+			free(ref->peer_ref);
+			free(ref);
+		} else
+			tail = &ref->next;
+	}
+
+	return ref_map;
+}
+
 static void query_refspecs_multiple(struct refspec *rs,
 				    struct refspec_item *query,
 				    struct string_list *results)
 {
-	int i;
+	int i, matched_negative = 0;
 	int find_src = !query->src;
+	struct string_list reversed = STRING_LIST_INIT_NODUP;
+	const char *needle = find_src ? query->dst : query->src;
+	char **result = find_src ? &query->src : &query->dst;
 
 	if (find_src && !query->dst)
 		BUG("query_refspecs_multiple: need either src or dst");
 
+	/*
+	 * If a ref matches any of the negative refspecs, then we should treat
+	 * it as not matching this query. Note that negative refspecs apply to
+	 * the source but we're checking only the destination. Reverse and
+	 * capture any pattern refspecs in order to see if the source would
+	 * have matched a negative refspec.
+	 */
+	for (i = 0; i < rs->nr; i++) {
+		struct refspec_item *refspec = &rs->items[i];
+		char *expn_name;
+
+		if (refspec->negative)
+			continue;
+
+		/* Note the reversal of src and dst */
+		if (refspec->pattern) {
+			const char *key = refspec->dst ?: refspec->src;
+			const char *value = refspec->src;
+
+			if (match_name_with_pattern(key, needle, value, &expn_name))
+				string_list_append_nodup(&reversed, expn_name);
+		} else {
+			if (!strcmp(needle, refspec->src))
+				string_list_append(&reversed, refspec->src);
+		}
+	}
+
+	for (i = 0; !matched_negative && i < reversed.nr; i++) {
+		if (omit_name_by_refspec(reversed.items[i].string, rs))
+			matched_negative = 1;
+	}
+
+	string_list_clear(&reversed, 0);
+
+	if (matched_negative)
+		return;
+
 	for (i = 0; i < rs->nr; i++) {
 		struct refspec_item *refspec = &rs->items[i];
 		const char *key = find_src ? refspec->dst : refspec->src;
 		const char *value = find_src ? refspec->src : refspec->dst;
-		const char *needle = find_src ? query->dst : query->src;
-		char **result = find_src ? &query->src : &query->dst;
 
-		if (!refspec->dst)
+		if (!refspec->dst || refspec->negative)
 			continue;
 		if (refspec->pattern) {
 			if (match_name_with_pattern(key, needle, value, result))
@@ -716,20 +792,58 @@  static void query_refspecs_multiple(struct refspec *rs,
 
 int query_refspecs(struct refspec *rs, struct refspec_item *query)
 {
-	int i;
+	int i, matched_negative = 0;
 	int find_src = !query->src;
+	struct string_list reversed = STRING_LIST_INIT_NODUP;
 	const char *needle = find_src ? query->dst : query->src;
 	char **result = find_src ? &query->src : &query->dst;
 
 	if (find_src && !query->dst)
 		BUG("query_refspecs: need either src or dst");
 
+	/*
+	 * If a ref matches any of the negative refspecs, then we should treat
+	 * it as not matching this query. Note that negative refspecs apply to
+	 * the source but we're checking only the destination. Reverse and
+	 * capture any pattern refspecs in order to see if the source would
+	 * have matched a negative refspec.
+	 */
+	for (i = 0; i < rs->nr; i++) {
+		struct refspec_item *refspec = &rs->items[i];
+		char *expn_name;
+
+		if (refspec->negative)
+			continue;
+
+		/* Note the reversal of src and dst */
+		if (refspec->pattern) {
+			const char *key = refspec->dst ?: refspec->src;
+			const char *value = refspec->src;
+
+			if (match_name_with_pattern(key, needle, value, &expn_name))
+				string_list_append_nodup(&reversed, expn_name);
+		} else {
+			if (!strcmp(needle, refspec->src))
+				string_list_append(&reversed, refspec->src);
+		}
+	}
+
+	for (i = 0; !matched_negative && i < reversed.nr; i++) {
+		if (omit_name_by_refspec(reversed.items[i].string, rs))
+			matched_negative = 1;
+	}
+
+	string_list_clear(&reversed, 0);
+
+	if (matched_negative)
+		return -1;
+
 	for (i = 0; i < rs->nr; i++) {
 		struct refspec_item *refspec = &rs->items[i];
 		const char *key = find_src ? refspec->dst : refspec->src;
 		const char *value = find_src ? refspec->src : refspec->dst;
 
-		if (!refspec->dst)
+		if (!refspec->dst || refspec->negative)
 			continue;
 		if (refspec->pattern) {
 			if (match_name_with_pattern(key, needle, value, result)) {
@@ -1058,7 +1172,7 @@  static int match_explicit(struct ref *src, struct ref *dst,
 	const char *dst_value = rs->dst;
 	char *dst_guess;
 
-	if (rs->pattern || rs->matching)
+	if (rs->pattern || rs->matching || rs->negative)
 		return 0;
 
 	matched_src = matched_dst = NULL;
@@ -1134,6 +1248,10 @@  static char *get_ref_match(const struct refspec *rs, const struct ref *ref,
 	int matching_refs = -1;
 	for (i = 0; i < rs->nr; i++) {
 		const struct refspec_item *item = &rs->items[i];
+
+		if (item->negative)
+			continue;
+
 		if (item->matching &&
 		    (matching_refs == -1 || item->force)) {
 			matching_refs = i;
@@ -1339,7 +1457,7 @@  int check_push_refs(struct ref *src, struct refspec *rs)
 	for (i = 0; i < rs->nr; i++) {
 		struct refspec_item *item = &rs->items[i];
 
-		if (item->pattern || item->matching)
+		if (item->pattern || item->matching || item->negative)
 			continue;
 
 		ret |= match_explicit_lhs(src, item, NULL, NULL);
@@ -1441,6 +1559,8 @@  int match_push_refs(struct ref *src, struct ref **dst,
 		string_list_clear(&src_ref_index, 0);
 	}
 
+	*dst = apply_negative_refspecs(*dst, rs);
+
 	if (errs)
 		return -1;
 	return 0;
@@ -1810,6 +1930,9 @@  int get_fetch_map(const struct ref *remote_refs,
 {
 	struct ref *ref_map, **rmp;
 
+	if (refspec->negative)
+		return 0;
+
 	if (refspec->pattern) {
 		ref_map = get_expanded_map(remote_refs, refspec);
 	} else {
diff --git a/remote.h b/remote.h
index 5e3ea5a26deb..104e75e0f74d 100644
--- a/remote.h
+++ b/remote.h
@@ -193,6 +193,12 @@  int resolve_remote_symref(struct ref *ref, struct ref *list);
  */
 struct ref *ref_remove_duplicates(struct ref *ref_map);
 
+/*
+ * Remove all entries in the input list which match any negative refspec in
+ * the refspec list.
+ */
+struct ref *apply_negative_refspecs(struct ref *ref_map, struct refspec *rs);
+
 int query_refspecs(struct refspec *rs, struct refspec_item *query);
 char *apply_refspecs(struct refspec *rs, const char *name);
 
@@ -205,7 +211,8 @@  void set_ref_status_for_push(struct ref *remote_refs, int send_mirror,
 /*
  * Given a list of the remote refs and the specification of things to
  * fetch, makes a (separate) list of the refs to fetch and the local
- * refs to store into.
+ * refs to store into. Note that negative refspecs are ignored here, and
+ * should be handled separately.
  *
  * *tail is the pointer to the tail pointer of the list of results
  * beforehand, and will be set to the tail pointer of the list of
diff --git a/t/t5582-fetch-negative-refspec.sh b/t/t5582-fetch-negative-refspec.sh
new file mode 100755
index 000000000000..83d08e513413
--- /dev/null
+++ b/t/t5582-fetch-negative-refspec.sh
@@ -0,0 +1,151 @@ 
+#!/bin/sh
+# Copyright (c) 2020, Jacob Keller.
+
+test_description='"git fetch" with negative refspecs.
+
+'
+
+. ./test-lib.sh
+
+test_expect_success setup '
+	echo >file original &&
+	git add file &&
+	git commit -a -m original
+'
+
+test_expect_success "clone and setup child repos" '
+	git clone . one &&
+	(
+		cd one &&
+		echo >file updated by one &&
+		git commit -a -m "updated by one" &&
+		git switch -c alternate &&
+		echo >file updated again by one &&
+		git commit -a -m "updated by one again" &&
+		git switch master
+	) &&
+	git clone . two &&
+	(
+		cd two &&
+		git config branch.master.remote one &&
+		git config remote.one.url ../one/.git/ &&
+		git config remote.one.fetch +refs/heads/*:refs/remotes/one/* &&
+		git config --add remote.one.fetch ^refs/heads/alternate
+	) &&
+	git clone . three
+'
+
+test_expect_success "fetch one" '
+	echo >file updated by origin &&
+	git commit -a -m "updated by origin" &&
+	(
+		cd two &&
+		test_must_fail git rev-parse --verify refs/remotes/one/alternate &&
+		git fetch one &&
+		test_must_fail git rev-parse --verify refs/remotes/one/alternate &&
+		git rev-parse --verify refs/remotes/one/master &&
+		mine=$(git rev-parse refs/remotes/one/master) &&
+		his=$(cd ../one && git rev-parse refs/heads/master) &&
+		test "z$mine" = "z$his"
+	)
+'
+
+test_expect_success "fetch with negative refspec on commandline" '
+	echo >file updated by origin again &&
+	git commit -a -m "updated by origin again" &&
+	(
+		cd three &&
+		alternate_in_one=$(cd ../one && git rev-parse refs/heads/alternate) &&
+		echo $alternate_in_one >expect &&
+		git fetch ../one/.git refs/heads/*:refs/remotes/one/* ^refs/heads/master &&
+		cut -f -1 .git/FETCH_HEAD >actual &&
+		test_cmp expect actual
+	)
+'
+
+test_expect_success "fetch with negative refspec avoids duplicate conflict" '
+	cd "$D" &&
+	(
+		cd one &&
+		git branch dups/a &&
+		git branch dups/b &&
+		git branch dups/c &&
+		git branch other/a &&
+		git rev-parse --verify refs/heads/other/a >../expect &&
+		git rev-parse --verify refs/heads/dups/b >>../expect &&
+		git rev-parse --verify refs/heads/dups/c >>../expect
+	) &&
+	(
+		cd three &&
+		git fetch ../one/.git ^refs/heads/dups/a refs/heads/dups/*:refs/dups/* refs/heads/other/a:refs/dups/a &&
+		git rev-parse --verify refs/dups/a >../actual &&
+		git rev-parse --verify refs/dups/b >>../actual &&
+		git rev-parse --verify refs/dups/c >>../actual
+	) &&
+	test_cmp expect actual
+'
+
+test_expect_success "push --prune with negative refspec" '
+	(
+		cd two &&
+		git branch prune/a &&
+		git branch prune/b &&
+		git branch prune/c &&
+		git push ../three refs/heads/prune/* &&
+		git branch -d prune/a &&
+		git branch -d prune/b &&
+		git push --prune ../three refs/heads/prune/* ^refs/heads/prune/b
+	) &&
+	(
+		cd three &&
+		test_write_lines b c >expect &&
+		git for-each-ref --format="%(refname:lstrip=3)" refs/heads/prune/ >actual &&
+		test_cmp expect actual
+	)
+'
+
+test_expect_success "push --prune with negative refspec apply to the destination" '
+	(
+		cd two &&
+		git branch ours/a &&
+		git branch ours/b &&
+		git branch ours/c &&
+		git push ../three refs/heads/ours/*:refs/heads/theirs/* &&
+		git branch -d ours/a &&
+		git branch -d ours/b &&
+		git push --prune ../three refs/heads/ours/*:refs/heads/theirs/* ^refs/heads/theirs/b
+	) &&
+	(
+		cd three &&
+		test_write_lines b c >expect &&
+		git for-each-ref --format="%(refname:lstrip=3)" refs/heads/theirs/ >actual &&
+		test_cmp expect actual
+	)
+'
+
+test_expect_success "fetch --prune with negative refspec" '
+	(
+		cd two &&
+		git branch fetch/a &&
+		git branch fetch/b &&
+		git branch fetch/c
+	) &&
+	(
+		cd three &&
+		git fetch ../two/.git refs/heads/fetch/*:refs/heads/copied/*
+	) &&
+	(
+		cd two &&
+		git branch -d fetch/a &&
+		git branch -d fetch/b
+	) &&
+	(
+		cd three &&
+		test_write_lines b c >expect &&
+		git fetch -v ../two/.git --prune refs/heads/fetch/*:refs/heads/copied/* ^refs/heads/fetch/b &&
+		git for-each-ref --format="%(refname:lstrip=3)" refs/heads/copied/ >actual &&
+		test_cmp expect actual
+	)
+'
+
+test_done