Skip to contentSkip to conversation

git-2.30.0-rc0 - Compile Fails on HPE NonStop

ANNOUNCE In review Randall S. Becker 4 participants 12 messages Dec 14, 2020 – Dec 16, 2020

lore
Share & tools
lore
#0 lore raw

2.30.0-rc0 fails to compile strmap.o, and probably others, since this is in a header file:

"git-compat-util.h", line 277: warning(1252):
missing return statement at end of non-void function "setitimer"

return strmap_remove(&map->map, str, 0);
^
"strmap.h", line 168: error(210):
a void function may not return a value

return strmap_remove(&set->map, str, 0);
^
"strmap.h", line 252: error(210):
a void function may not return a value

Aside from inlining bodies, this should not have compiled on any platform:

static inline void strset_remove(struct strset *set, const char *str)

{
        return strmap_remove(&set->map, str, 0);
}

What is really intended here?

Sorry,
Randall

-- Brief whoami:
NonStop developer since approximately 211288444200000000 UNIX developer since approximately 421664400
-- In my real life, I talk too much.

#1 lore raw
Quoted 5 lines · Randall S. Becker

2.30.0-rc0 fails to compile strmap.o, and probably others, since this is in a header file:

"git-compat-util.h", line 277: warning(1252):
missing return statement at end of non-void function "setitimer"

Thanks for reporting. I guess nobody without setitimer bothered to test, since 15b52a44 (compat-util: type-check parameters of no-op replacement functions, 2020-08-06) which was a tad before 2.29 was tagged.

git-compat-util.h +2 -1
index 7d509c5022..58cd0761be 100644
@@ -273,7 +273,8 @@ struct itimerval {
 
 #ifdef NO_SETITIMER
 static inline int setitimer(int which, const struct itimerval *value, struct itimerval *newvalue) {
-	; /* nothing */
+	errno = ENOSYS;
+	return -1; /* not implemented */
 }
 #endif
 

Alternatively we could pretend that the call always succeeds by without touching errno and returning 0. That might be safer, but I dunno which one we want, and I do not have a system affected by the choice.

Quoted 9 lines

return strmap_remove(&map->map, str, 0);
^
"strmap.h", line 168: error(210):
a void function may not return a value

return strmap_remove(&set->map, str, 0);
^
"strmap.h", line 252: error(210):
a void function may not return a value

This is a GNU extension biting us, perhaps?

Apparently these came from 4fa1d501 (strmap: add functions facilitating use as a string->int map, 2020-11-05).

Quoted 8 lines

Aside from inlining bodies, this should not have compiled on any platform:

static inline void strset_remove(struct strset *set, const char *str)

{
        return strmap_remove(&set->map, str, 0);
}

What is really intended here?

I think we should just drop "return"; a void function should be called in void context without requiring a value, even if that return expects no value.

strmap.h +2 -2
index c4c104411b..1e152d832d 100644
@@ -165,7 +165,7 @@ static inline int strintmap_contains(struct strintmap *map, const char *str)
 
 static inline void strintmap_remove(struct strintmap *map, const char *str)
 {
-	return strmap_remove(&map->map, str, 0);
+	strmap_remove(&map->map, str, 0);
 }
 
 static inline int strintmap_empty(struct strintmap *map)
@@ -249,7 +249,7 @@ static inline int strset_contains(struct strset *set, const char *str)
 
 static inline void strset_remove(struct strset *set, const char *str)
 {
-	return strmap_remove(&set->map, str, 0);
+	strmap_remove(&set->map, str, 0);
 }
 
 static inline int strset_empty(struct strset *set)
Quoted 2 lines

Sorry,
Randall

Thanks.

#2 lore raw
Quoted 14 lines · Junio C Hamano
git-compat-util.h +2 -1
index 7d509c5022..58cd0761be 100644
@@ -273,7 +273,8 @@ struct itimerval {
 
 #ifdef NO_SETITIMER
 static inline int setitimer(int which, const struct itimerval *value, struct itimerval *newvalue) {
-	; /* nothing */
+	errno = ENOSYS;
+	return -1; /* not implemented */
 }
 #endif
 
Quoted 4 lines

Alternatively we could pretend that the call always succeeds by without touching errno and returning 0. That might be safer, but I dunno which one we want, and I do not have a system affected by the choice.

I think this is a sensible choice. Before the conversion to an inline, the code was removed entirely! So anybody checking the return value would have seen an error, and we do not have to worry much about breaking them.

For new callers, anybody checking the return value would probably appreciate the warning that support for the function is optional (OTOH, they would probably not find out themselves, but rather when Randall tells them ;) ).

It would be nice to have a way to warn them even on platforms that have setitimer(), but I can't think of an easy way.

Quoted 7 lines
Quoted 7 lines

Aside from inlining bodies, this should not have compiled on any platform:

static inline void strset_remove(struct strset *set, const char *str)

{
        return strmap_remove(&set->map, str, 0);
}
Quoted 5 lines
Quoted 1 line

What is really intended here?

I think we should just drop "return"; a void function should be called in void context without requiring a value, even if that return expects no value.

Yeah, I think that is right. I checked earlier iterations of the series to see if perhaps strmap_remove() had previously returned a value, but it never did in any on the list.

strmap.h static inline int strintmap_contains(struct strintmap *map, const char *str) · line 169

Quoted 11 lines
strmap.h +1 -1
index c4c104411b..1e152d832d 100644
@@ -165,7 +165,7 @@ static inline int strintmap_contains(struct strintmap *map, const char *str)
 
 static inline void strintmap_remove(struct strintmap *map, const char *str)
 {
-	return strmap_remove(&map->map, str, 0);
+	strmap_remove(&map->map, str, 0);
 }

So yeah, I think that is the right fix. +cc Elijah for any other insight.

-Peff

#3 lore raw
Quoted 26 lines · Peff

git-compat-util.h struct itimerval { · line 275

Quoted 8 lines · Junio C Hamano
w/git-compat-util.h index renamed from c/git-compat-util.h +0 -0

7d509c5022..58cd0761be 100644

git-compat-util.h +0 -0
@@ -273,7 +273,8 @@ struct itimerval {
 
 #ifdef NO_SETITIMER
 static inline int setitimer(int which, const struct itimerval *value, struct

itimerval *newvalue) {

Quoted 10 lines
git-compat-util.h +2 -1
-	; /* nothing */
+	errno = ENOSYS;
+	return -1; /* not implemented */
 }
 #endif
 

Alternatively we could pretend that the call always succeeds by without touching errno and returning 0. That might be safer, but I dunno which one we want, and I do not have a system affected by the choice.

I think this is a sensible choice. Before the conversion to an inline, the code was removed entirely! So anybody checking the return value would have seen an error, and we do not have to worry much about breaking them.

Quoted 7 lines

For new callers, anybody checking the return value would probably appreciate the warning that support for the function is optional (OTOH, they would probably not find out themselves, but rather when Randall tells them ;) ).

It would be nice to have a way to warn them even on platforms that have setitimer(), but I can't think of an easy way.

Strangely, we do have setitimer() on NonStop. I'm not sure how we get into this situation, unless it's not being detected correctly.

Quoted 24 lines
Quoted 1 line
Quoted 1 line

Aside from inlining bodies, this should not have compiled on any

platform:

Quoted 10 lines
Quoted 6 lines

static inline void strset_remove(struct strset *set, const char

*str) {
        return strmap_remove(&set->map, str, 0); }

What is really intended here?

I think we should just drop "return"; a void function should be called in void context without requiring a value, even if that return expects no value.

Yeah, I think that is right. I checked earlier iterations of the series to see if perhaps strmap_remove() had previously returned a value, but it never did in any on the list.

Quoted 7 lines
strmap.h +0 -0
index c4c104411b..1e152d832d 100644
@@ -165,7 +165,7 @@ static inline int strintmap_contains(struct
 strintmap *map, const char *str)
 
Quoted 7 lines

strmap.h static inline int strintmap_contains(struct strintmap *map, const char *str) · line 169

Quoted 5 lines
strmap.h +1 -1
 static inline void strintmap_remove(struct strintmap *map, const char
 *str)  {
-	return strmap_remove(&map->map, str, 0);
+	strmap_remove(&map->map, str, 0);
 }

So yeah, I think that is the right fix. +cc Elijah for any other insight.

FYI: We successfully built 2.29.2 without issues. I'm not sure how this all happened but seems like it was at 1201eb628a.

Thanks,
Randall

#4 lore raw
In reply to Jeff King
Quoted 58 lines · Jeff King
Quoted 14 lines · Junio C Hamano
git-compat-util.h +2 -1
index 7d509c5022..58cd0761be 100644
@@ -273,7 +273,8 @@ struct itimerval {
 
 #ifdef NO_SETITIMER
 static inline int setitimer(int which, const struct itimerval *value, struct itimerval *newvalue) {
-     ; /* nothing */
+     errno = ENOSYS;
+     return -1; /* not implemented */
 }
 #endif
 
Quoted 4 lines

Alternatively we could pretend that the call always succeeds by without touching errno and returning 0. That might be safer, but I dunno which one we want, and I do not have a system affected by the choice.

I think this is a sensible choice. Before the conversion to an inline, the code was removed entirely! So anybody checking the return value would have seen an error, and we do not have to worry much about breaking them.

For new callers, anybody checking the return value would probably appreciate the warning that support for the function is optional (OTOH, they would probably not find out themselves, but rather when Randall tells them ;) ).

It would be nice to have a way to warn them even on platforms that have setitimer(), but I can't think of an easy way.

Quoted 7 lines
Quoted 7 lines

Aside from inlining bodies, this should not have compiled on any platform:

static inline void strset_remove(struct strset *set, const char *str)

{
        return strmap_remove(&set->map, str, 0);
}
Quoted 5 lines
Quoted 1 line

What is really intended here?

I think we should just drop "return"; a void function should be called in void context without requiring a value, even if that return expects no value.

Yeah, I think that is right. I checked earlier iterations of the series to see if perhaps strmap_remove() had previously returned a value, but it never did in any on the list.

Quoted 6 lines
strmap.h +0 -0
index c4c104411b..1e152d832d 100644
@@ -165,7 +165,7 @@ static inline int strintmap_contains(struct strintmap *map, const char *str)
 
Quoted 8 lines

strmap.h static inline int strintmap_contains(struct strintmap *map, const char *str) · line 169

Quoted 5 lines
strmap.h +1 -1
 static inline void strintmap_remove(struct strintmap *map, const char *str)
 {
-     return strmap_remove(&map->map, str, 0);
+     strmap_remove(&map->map, str, 0);
 }

So yeah, I think that is the right fix. +cc Elijah for any other insight.

Doh, sorry for the bug. Yeah, that's the exact fix I'd make. Junio, do you want to just use the changes you've already got, or would you like me to submit a patch removing these two 'return's?

#6 lore raw
In reply to Elijah Newren
Quoted 3 lines · Elijah Newren

Doh, sorry for the bug. Yeah, that's the exact fix I'd make. Junio, do you want to just use the changes you've already got, or would you like me to submit a patch removing these two 'return's?

Nah, the other half of the reported issue was mine, so let me see if I can quickly whip up a two-patch series.

Thanks.

#7 lore raw

PATCH 1/1 strmap: make callers of strmap_remove() to call it in void context

Two "static inline" functions, both of which return void, call strmap_remove() and tries to return the value it returns as their return value, which is just bogus, as strmap_remove() returns void itself. Call it in the void context and fall-thru the control to the end instead.

---
 strmap.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
strmap.h +2 -2
index c4c104411b..1e152d832d 100644
@@ -165,7 +165,7 @@ static inline int strintmap_contains(struct strintmap *map, const char *str)
 
 static inline void strintmap_remove(struct strintmap *map, const char *str)
 {
-	return strmap_remove(&map->map, str, 0);
+	strmap_remove(&map->map, str, 0);
 }
 
 static inline int strintmap_empty(struct strintmap *map)
@@ -249,7 +249,7 @@ static inline int strset_contains(struct strset *set, const char *str)
 
 static inline void strset_remove(struct strset *set, const char *str)
 {
-	return strmap_remove(&set->map, str, 0);
+	strmap_remove(&set->map, str, 0);
 }
 
 static inline int strset_empty(struct strset *set)
Signature · 1 line
2.30.0-rc0-186-g20447144ec
#8 lore raw

RE: [PATCH] strmap: make callers of strmap_remove() to call it in void context

Quoted 7 lines

-----Original Message-----
From: Junio C Hamano <[email protected]>
Sent: December 15, 2020 4:26 PM
To: Elijah Newren <[email protected]>; Randall S. Becker <[email protected]>
Cc: Jeff King <[email protected]>; Git Mailing List <[email protected]>
Subject: [PATCH] strmap: make callers of strmap_remove() to call it in

void

Quoted 4 lines

context

Two "static inline" functions, both of which return void, call strmap_remove() and tries to return the value it returns as their return

value,

Quoted 1 line

which is just bogus, as strmap_remove() returns void itself. Call it in

the void

Quoted 13 lines

context and fall-thru the control to the end instead.

---
 strmap.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
strmap.h +0 -0
index c4c104411b..1e152d832d 100644

strmap.h static inline int strintmap_contains(struct strintmap · line 167

Quoted 3 lines
strmap.h +0 -0
@@ -165,7 +165,7 @@ static inline int strintmap_contains(struct strintmap
 *map, const char *str)
 
 static inline void strintmap_remove(struct strintmap *map, const char

*str) {

Quoted 14 lines
strmap.h +1 -1
-	return strmap_remove(&map->map, str, 0);
+	strmap_remove(&map->map, str, 0);
 }
 
 static inline int strintmap_empty(struct strintmap *map) @@ -249,7 +249,7

@@ static inline int strset_contains(struct strset *set, const char *str)

static inline void strset_remove(struct strset *set, const char *str) {

strmap.h +1 -2
-	return strmap_remove(&set->map, str, 0);
+	strmap_remove(&set->map, str, 0);
 }
 
 static inline int strset_empty(struct strset *set)
--
Quoted 1 line

2.30.0-rc0-186-g20447144ec

Looks good from here.

Regards,
Randall

#9 lore raw

PATCH 1/1 compat-util: pretend that stub setitimer() always succeeds

In reply to Junio C Hamano

When 15b52a44 (compat-util: type-check parameters of no-op replacement functions, 2020-08-06) turned a handful of no-op C-preprocessor macros into static inline functions to give the callers a better type checking for their parameters, it forgot to return anything from the stubbed out setitimer() function, even though the function was defined to return an int just like the real thing.

Since the original C-preprocessor macro implementation was to just turn the call to the function an empty statement, we know that the existing callers do not check the return value from it, and it does not matter what value we return. But it is safer to pretend that the call succeeded by returning 0 than making it fail by returning -1 and clobbering errno with some value.

---
 git-compat-util.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
git-compat-util.h +1 -1
index 7a0fb7a045..421c90b5d8 100644
@@ -273,7 +273,7 @@ struct itimerval {
 
 #ifdef NO_SETITIMER
 static inline int setitimer(int which, const struct itimerval *value, struct itimerval *newvalue) {
-	; /* nothing */
+	return 0; /* pretend success */
 }
 #endif
 
Signature · 1 line
2.30.0-rc0-186-g20447144ec
#10 lore raw
In reply to Junio C Hamano
Quoted 8 lines · Junio C Hamano
Quoted 3 lines · Elijah Newren

Doh, sorry for the bug. Yeah, that's the exact fix I'd make. Junio, do you want to just use the changes you've already got, or would you like me to submit a patch removing these two 'return's?

Nah, the other half of the reported issue was mine, so let me see if I can quickly whip up a two-patch series.

Thanks, both look OK to me. I probably would have kept "return -1" in the setitimer one, but I doubt it really matters much either way.

-Peff

#11 lore raw
Quoted 8 lines · Jeff King
Quoted 6 lines · Junio C Hamano
Quoted 3 lines · Elijah Newren

Doh, sorry for the bug. Yeah, that's the exact fix I'd make. Junio, do you want to just use the changes you've already got, or would you like me to submit a patch removing these two 'return's?

Quoted 5 lines
Quoted 2 lines

Nah, the other half of the reported issue was mine, so let me see if I can quickly whip up a two-patch series.

Thanks, both look OK to me. I probably would have kept "return -1" in the setitimer one, but I doubt it really matters much either way.

Yeah, we know that the return value does not matter as nobody looks at it. I just was a bit hesitant to clobber errno with ENOSYS.