submodule.h: use a named enum for RECURSE_SUBMODULES_*
PATCH 1/1 submodule.h: use a named enum for RECURSE_SUBMODULES_*
From: Philippe Blain <[email protected]>
Using a named enum allows casting an integer to the enum type in both GDB and LLDB:
(gdb) p (enum diff_submodule_format) options->submodule_format $1 = DIFF_SUBMODULE_LOG
(lldb) p (diff_submodule_format) options->submodule_format (diff_submodule_format) $1 = DIFF_SUBMODULE_LOG
In LLDB, it's also required to cast in the reversed direction, i.e. cast an enum constant into its corresponding integer:
(lldb) p (int) diff_submodule_format::DIFF_SUBMODULE_SHORT (int) $0 = 0
Name the enum listing the different RECURSE_SUBMODULES_* modes, to make debugging easier.
---
submodule.h: use a named enum for RECURSE_SUBMODULES_*
Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-1111%2Fphil-blain%2Fsubmodule-recurse-enum-v1
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-1111/phil-blain/submodule-recurse-enum-v1
Pull-Request: https://github.com/gitgitgadget/git/pull/1111
submodule.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
submodule.h +1 -1
@@ -13,7 +13,7 @@ struct repository;struct string_list ;struct strbuf ; -enum { +enum submodule_recurse_mode {RECURSE_SUBMODULES_ONLY =-5 ,RECURSE_SUBMODULES_CHECK =-4 ,RECURSE_SUBMODULES_ERROR =-3 ,
base-commit: 2ae0a9cb8298185a94e5998086f380a355dd8907
Signature · 1 line
gitgitgadget
Quoted 7 lines · Philippe Blain via GitGitGadget
From: Philippe Blain <[email protected]>
Using a named enum allows casting an integer to the enum type in both GDB and LLDB:
(gdb) p (enum
diff_submodule_format)options->submodule_format$1 =DIFF_SUBMODULE_LOG
Hmph, this was somewhat unexpected and quite disappointing.
Because that's not what those "Let's move away from #define'd constants and use more enums" said when they sold their "enum" to us. In the diff_options struct, the .submodule_format member is of type enum already, so, if we trust what they told us when they sold their enums, "p options->submodule_format" should be enough to give us "DIFF_SUBMODULE_LOG", not "1", as its output. Do you really need the cast in the above example?
Quoted 2 lines
Name the enum listing the different RECURSE_SUBMODULES_* modes, to make debugging easier.
Even though this patch may be a good single step in the right direction, until it is _used_ to declare a variable or a member of a struct of that enum type, it probably is not useful as much as it could be. The user needs to know which enum is stored in that "int" variable or member the user is interested in to cast it to.
That is, many changes like this one.
builtin/pull.c +1 -1
@@ -71,7 +71,7 @@ static const char * const pull_usage[] = {/* Shared options */ static int opt_verbosity;static char *opt_progress; -static int recurse_submodules =RECURSE_SUBMODULES_DEFAULT ; +static enum submodule_recurse_mode recurse_submodules =RECURSE_SUBMODULES_DEFAULT ;/* Options passed to git-merge or git-rebase */ static enum rebase_type opt_rebase =-1 ;
Hi Junio,
Le 2022-01-05 à 16:20, Junio C Hamano a écrit :
Quoted 12 lines
Quoted 7 lines · Philippe Blain via GitGitGadget
From: Philippe Blain <[email protected]>
Using a named enum allows casting an integer to the enum type in both GDB and LLDB:
(gdb) p (enum
diff_submodule_format)options->submodule_format$1 =DIFF_SUBMODULE_LOGHmph, this was somewhat unexpected and quite disappointing.
Quoted 7 lines
Because that's not what those "Let's move away from #define'd constants and use more enums" said when they sold their "enum" to us. In the
diff_optionsstruct, the .submodule_format member is of type enum already, so, if we trust what they told us when they sold their enums, "poptions->submodule_format" should be enough to give us "DIFF_SUBMODULE_LOG", not "1", as its output. Do you really need the cast in the above example?
Yes, you are right that my example does not reflect what I'm saying, since options->submodule_format is not an int. I checked and indeed we do not need any cast to get "DIFF_SUBMODULE_LOG". We do need it when dealing with int's, which is not the case here. I'll try to find a better example.
Quoted 9 lines
Quoted 3 lines
Name the enum listing the different RECURSE_SUBMODULES_* modes, to make debugging easier.
Even though this patch may be a good single step in the right direction, until it is _used_ to declare a variable or a member of a struct of that enum type, it probably is not useful as much as it could be. The user needs to know which enum is stored in that "int" variable or member the user is interested in to cast it to.
Yes, that's true. But when I came across that, I was in a place of the code where some int was compared with a constant in this enum, RECURSE_SUBMODULES_something. So it would have been easy to check where the enum is declared, learn its name and use it to cast the int to the enum type. That's the kind of situation I have in mind.
Quoted 9 lines
That is, many changes like this one.
builtin/pull.c +0
-0
builtin/pull.c static const char * const pull_usage[] = { · line 77
Quoted 8 lines
builtin/pull.c +1
-1@@ -71,7 +71,7 @@ static const char * const pull_usage[] = {/* Shared options */ static int opt_verbosity;static char *opt_progress; -static int recurse_submodules =RECURSE_SUBMODULES_DEFAULT ; +static enum submodule_recurse_mode recurse_submodules =RECURSE_SUBMODULES_DEFAULT ;/* Options passed to git-merge or git-rebase */ static enum rebase_type opt_rebase =-1 ;
Yes, this is a parallel effort that could be done, I agree, but my patch was meant to help in the mean time.
Thanks,
Philippe.
Quoted 14 lines · Philippe Blain
Quoted 14 lines
That is, many changes like this one.
builtin/pull.c +1
-1@@ -71,7 +71,7 @@ static const char * const pull_usage[] = {/* Shared options */ static int opt_verbosity;static char *opt_progress; -static int recurse_submodules =RECURSE_SUBMODULES_DEFAULT ; +static enum submodule_recurse_mode recurse_submodules =RECURSE_SUBMODULES_DEFAULT ;
Quoted 6 lines
builtin/pull.c static const char * const pull_usage[] = { · line 77
Quoted 3 lines
builtin/pull.c +0
-0/* Options passed to git-merge or git-rebase */ static enum rebase_type opt_rebase =-1 ;Yes, this is a parallel effort that could be done, I agree, but my patch was meant to help in the mean time.
There are quite a few sites that could use this
s/int/enum submodule_recurse_mode change. I suppose one _could_ change all of them at once, but that seems cumbersome to review and prone to conflict.
So that this isn't debugger-only, I'd be happy with at least one change (perhaps the one that inspired you to name the enum in the first place), and making the other changes when it makes sense, e.g. I can do this for the fetch machinery while I work on enhancements to fetch --recurse-submodules.
Quoted 36 lines · Junio C Hamano
Quoted 7 lines · Philippe Blain via GitGitGadget
From: Philippe Blain <[email protected]>
Using a named enum allows casting an integer to the enum type in both GDB and LLDB:
(gdb) p (enum
diff_submodule_format)options->submodule_format$1 =DIFF_SUBMODULE_LOGHmph, this was somewhat unexpected and quite disappointing.
Because that's not what those "Let's move away from #define'd constants and use more enums" said when they sold their "enum" to us. In the
diff_optionsstruct, the .submodule_format member is of type enum already, so, if we trust what they told us when they sold their enums, "poptions->submodule_format" should be enough to give us "DIFF_SUBMODULE_LOG", not "1", as its output. Do you really need the cast in the above example?Quoted 2 lines
Name the enum listing the different RECURSE_SUBMODULES_* modes, to make debugging easier.
Even though this patch may be a good single step in the right direction, until it is _used_ to declare a variable or a member of a struct of that enum type, it probably is not useful as much as it could be. The user needs to know which enum is stored in that "int" variable or member the user is interested in to cast it to.
That is, many changes like this one.
builtin/pull.c +0
-0
builtin/pull.c static const char * const pull_usage[] = { · line 77
Quoted 8 lines
builtin/pull.c +1
-1@@ -71,7 +71,7 @@ static const char * const pull_usage[] = {/* Shared options */ static int opt_verbosity;static char *opt_progress; -static int recurse_submodules =RECURSE_SUBMODULES_DEFAULT ; +static enum submodule_recurse_mode recurse_submodules =RECURSE_SUBMODULES_DEFAULT ;/* Options passed to git-merge or git-rebase */ static enum rebase_type opt_rebase =-1 ;
Your comment on the use of RECURSE_SUBMODULES_DEFAULT elsewhere [1] reminded me of how odd this enum actually is.
* submodule_recurse_mode is used almost exclusively by fetch and push because they are the only commands that accept anything other than true/false.
* however, it is also used by submodule.c:config_update_recurse_submodules to store true/false (it don't even use RECURSE_SUBMODULES_DEFAULT!)
i.e. I suspect that the enum is only relevant for fetch/push, but its values for _ON and _OFF have been co-opted for other things.
This patch and the suggestion of s/int/enum submodule_recurse_mode makes sense, though I think we can take this a bit further in some follow-up work:
If I am correct in my suspicion earlier, then submodule_recurse_mode can be made even more specific, like "submodule_transport_mode", and all non-transport related uses can be replaced with int.
If I am wrong and there are some legitimate uses for submodule_recurse_mode that I have missed, it might still be worthwhile to separate the different uses of the enum so that it doesn't end up overloaded.
Quoted 16 lines · Philippe Blain via GitGitGadget
From: Philippe Blain <[email protected]>
Using a named enum allows casting an integer to the enum type in both GDB and LLDB:
$ gdb -q -ex 'b wt-status.c:44' -ex r --args ./git status (gdb) p (enum color_wt_status) slot $1 = WT_STATUS_ONBRANCH$ lldb -o 'b wt-status.c:44' -o r -- ./git status (lldb) p (color_wt_status) slot (color_wt_status) $0 = WT_STATUS_ONBRANCHIn LLDB, it's also required to cast in the reversed direction, i.e. cast an enum constant into its corresponding integer:
Quoted 8 lines
(lldb) p (int)
color_wt_status::WT_STATUS_ONBRANCH(int) $1 = 8Name the enum listing the different RECURSE_SUBMODULES_* modes, to make debugging easier. For example, when stepping through a part of the code where an int is compared with a constant in this enum, it allows casting the int to the enum type or vice-versa, after quickly checking where the enum constant is declared and learning the enum name.
Makes sense, and besides just making debugging easier, this would also make code easier to read once we use the enum in more places (instead of just int).
Quoted 3 lines
As to not make this patch a debug-only change, convert the '
fetch_recurse' member of 'struct submodule' to use the newly named enum.
[...]
Quoted 22 lines
submodule-config.h | 2 +- submodule.h | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-)submodule-config.h +1
-1@@ -37,7 +37,7 @@ struct submodule {const char *path;const char *name;const char *url; -int fetch_recurse; +enum submodule_recurse_mode fetch_recurse;const char *ignore;const char *branch;struct submodule_update_strategy update_strategy;submodule.h +0
-0
submodule.h struct repository; · line 19
Quoted 8 lines
submodule.h +1
-1@@ -13,7 +13,7 @@ struct repository;struct string_list ;struct strbuf ; -enum { +enum submodule_recurse_mode {RECURSE_SUBMODULES_ONLY =-5 ,RECURSE_SUBMODULES_CHECK =-4 ,RECURSE_SUBMODULES_ERROR =-3 ,
Like Junio, I think it would be nice to use the enum in more places, but frankly there are so many sites that would need to change that I don't think it's reasonable for one person to change them all.
So I'm happy to take this first step and do the rest of the refactoring incrementally. I still think the enum's values are too disjointed and should eventually be broken up, but that's a refactoring for later.
PATCH v2 1/1 submodule.h: use a named enum for RECURSE_SUBMODULES_*
In reply to Philippe Blain via GitGitGadgetFrom: Philippe Blain <[email protected]>
Using a named enum allows casting an integer to the enum type in both GDB and LLDB:
$ gdb -q -ex 'b wt-status.c:44' -ex r --args ./git status
(gdb) p (enum color_wt_status) slot
$1 = WT_STATUS_ONBRANCH $ lldb -o 'b wt-status.c:44' -o r -- ./git status
(lldb) p (color_wt_status) slot
(color_wt_status) $0 = WT_STATUS_ONBRANCHIn LLDB, it's also required to cast in the reversed direction, i.e. cast an enum constant into its corresponding integer:
(lldb) p (int) color_wt_status::WT_STATUS_ONBRANCH (int) $1 = 8
Name the enum listing the different RECURSE_SUBMODULES_* modes, to make debugging easier. For example, when stepping through a part of the code where an int is compared with a constant in this enum, it allows casting the int to the enum type or vice-versa, after quickly checking where the enum constant is declared and learning the enum name.
As to not make this patch a debug-only change, convert the 'fetch_recurse' member of 'struct submodule' to use the newly named enum.
---
submodule.h: use a named enum for RECURSE_SUBMODULES_*
Changes since v1:
* converted one user of the enum from an 'int' to the new enum type, so the patch is not debug-only, as suggested by Glen.
* updated the commit message to use an 'int' as example of a local variable being compared with an enum constant, instead of using a struct member which is already of enum type, as pointed out by Junio
* added a little bit more explanation in the commit message as to how this patch can help when debugging.
Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-1111%2Fphil-blain%2Fsubmodule-recurse-enum-v2
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-1111/phil-blain/submodule-recurse-enum-v2
Pull-Request: https://github.com/gitgitgadget/git/pull/1111
Range-diff vs v1:
1: 4c787d4054e ! 1: e0b211b88f5 submodule.h: use a named enum for RECURSE_SUBMODULES_*
@@ Commit message
Using a named enum allows casting an integer to the enum type in both GDB and LLDB:
- (gdb) p (enum diff_submodule_format) options->submodule_format
- $1 = DIFF_SUBMODULE_LOG
+ $ gdb -q -ex 'b wt-status.c:44' -ex r --args ./git status
+ (gdb) p (enum color_wt_status) slot
+ $1 = WT_STATUS_ONBRANCH
- (lldb) p (diff_submodule_format) options->submodule_format
- (diff_submodule_format) $1 = DIFF_SUBMODULE_LOG
+ $ lldb -o 'b wt-status.c:44' -o r -- ./git status
+ (lldb) p (color_wt_status) slot
+ (color_wt_status) $0 = WT_STATUS_ONBRANCH
In LLDB, it's also required to cast in the reversed direction, i.e. cast an enum constant into its corresponding integer:
- (lldb) p (int) diff_submodule_format::DIFF_SUBMODULE_SHORT
- (int) $0 = 0
+ (lldb) p (int) color_wt_status::WT_STATUS_ONBRANCH
+ (int) $1 = 8
Name the enum listing the different RECURSE_SUBMODULES_* modes, to make
- debugging easier.
+ debugging easier. For example, when stepping through a part of the code
+ where an int is compared with a constant in this enum, it allows casting
+ the int to the enum type or vice-versa, after quickly checking where the
+ enum constant is declared and learning the enum name.
+
+ As to not make this patch a debug-only change, convert the
+ 'fetch_recurse' member of 'struct submodule' to use the newly named
+ enum.
+ ## submodule-config.h ##
+@@ submodule-config.h: struct submodule {
+ const char *path;
+ const char *name;
+ const char *url;
+- int fetch_recurse;
++ enum submodule_recurse_mode fetch_recurse;
+ const char *ignore;
+ const char *branch;
+ struct submodule_update_strategy update_strategy;
+
## submodule.h ##
@@ submodule.h: struct repository;
struct string_list;
submodule-config.h | 2 +- submodule.h | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-)
submodule-config.h +1 -1
@@ -37,7 +37,7 @@ struct submodule {const char *path;const char *name;const char *url; -int fetch_recurse; +enum submodule_recurse_mode fetch_recurse;const char *ignore;const char *branch;struct submodule_update_strategy update_strategy;
submodule.h +1 -1
@@ -13,7 +13,7 @@ struct repository;struct string_list ;struct strbuf ; -enum { +enum submodule_recurse_mode {RECURSE_SUBMODULES_ONLY =-5 ,RECURSE_SUBMODULES_CHECK =-4 ,RECURSE_SUBMODULES_ERROR =-3 ,
base-commit: 2ae0a9cb8298185a94e5998086f380a355dd8907
Signature · 1 line
gitgitgadget