If a tristate choice has another choice dependers, kconfig produce
wrong value.
Example:
choice
prompt "choice list AB"
config A
tristate "a"
config B
tristate "b"
endchoice
choice
prompt "choice list CD"
depends on A
config C
bool "c"
config D
bool "d"
endchoice
kconfig outputs CONFIG_A=m from defconfig with CONFIG_A=y.
This happens if the choice list CD depends on A.
Also, if defconfig contains CONFIG_A=y and "# CONFIG_B is not set",
output is correct.
I found commit a64b44ea ("kconfig: fix tristate choice with minimal
config"), but this problem still exists on v3.3-rc1. More fix is
needed for choice with dependers?
---
Atsushi Nemoto
(added Arnaud to CC)
On 7.2.2012 01:16, Atsushi Nemoto wrote:
> If a tristate choice has another choice dependers, kconfig produce
> wrong value.
>
> Example:
>
> choice
> prompt "choice list AB"
> config A
> tristate "a"
> config B
> tristate "b"
> endchoice
> choice
> prompt "choice list CD"
> depends on A
> config C
> bool "c"
> config D
> bool "d"
> endchoice
>
> kconfig outputs CONFIG_A=m from defconfig with CONFIG_A=y.
>
> This happens if the choice list CD depends on A.
> Also, if defconfig contains CONFIG_A=y and "# CONFIG_B is not set",
> output is correct.
>
> I found commit a64b44ea ("kconfig: fix tristate choice with minimal
> config"), but this problem still exists on v3.3-rc1. More fix is
> needed for choice with dependers?
>
> ---
> Atsushi Nemoto
Hi,
On Mon, Feb 6, 2012 at 7:16 PM, Atsushi Nemoto <[email protected]> wrote:
> If a tristate choice has another choice dependers, kconfig produce
> wrong value.
>
> Example:
>
> choice
> ? ? ? ?prompt "choice list AB"
> config A
> ? ? ? ?tristate "a"
> config B
> ? ? ? ?tristate "b"
> endchoice
> choice
> ? ? ? ?prompt "choice list CD"
> ? ? ? ?depends on A
> config C
> ? ? ? ?bool "c"
> config D
> ? ? ? ?bool "d"
> endchoice
>
> kconfig outputs CONFIG_A=m from defconfig with CONFIG_A=y.
>
> This happens if the choice list CD depends on A.
> Also, if defconfig contains CONFIG_A=y and "# CONFIG_B is not set",
> output is correct.
>
confirmed.
> I found commit a64b44ea ("kconfig: fix tristate choice with minimal
> config"), but this problem still exists on v3.3-rc1. ?More fix is
> needed for choice with dependers?
>
it would seem so.
- Arnaud
Hi,
On Thu, Feb 9, 2012 at 12:08 AM, Arnaud Lacombe <[email protected]> wrote:
> Hi,
>
> On Mon, Feb 6, 2012 at 7:16 PM, Atsushi Nemoto <[email protected]> wrote:
>> If a tristate choice has another choice dependers, kconfig produce
>> wrong value.
>>
>> Example:
>>
>> choice
>> ? ? ? ?prompt "choice list AB"
>> config A
>> ? ? ? ?tristate "a"
>> config B
>> ? ? ? ?tristate "b"
>> endchoice
>> choice
>> ? ? ? ?prompt "choice list CD"
>> ? ? ? ?depends on A
>> config C
>> ? ? ? ?bool "c"
>> config D
>> ? ? ? ?bool "d"
>> endchoice
>>
>> kconfig outputs CONFIG_A=m from defconfig with CONFIG_A=y.
>>
>> This happens if the choice list CD depends on A.
>> Also, if defconfig contains CONFIG_A=y and "# CONFIG_B is not set",
>> output is correct.
>>
> confirmed.
>
actually, not that much. Your reduced testcase is missing a MODULES
symbol to have modules enabled and have them show up in the final
`.config'. If you add:
config MODULES
boolean
default y
then you end up with CONFIG_A=m in the `.config'.
Now, I may have missed something.
Do you have a situation where a MODULES symbol is present, enabled,
but CONFIG_A=y still appear from CONFIG_A=m in the defconfig ?
Thanks,
- Arnaud
>> I found commit a64b44ea ("kconfig: fix tristate choice with minimal
>> config"), but this problem still exists on v3.3-rc1. ?More fix is
>> needed for choice with dependers?
>>
> it would seem so.
>
> ?- Arnaud
On Thu, 9 Feb 2012 00:26:25 -0500, Arnaud Lacombe <[email protected]> wrote:
>> confirmed.
>>
> actually, not that much. Your reduced testcase is missing a MODULES
> symbol to have modules enabled and have them show up in the final
> `.config'. If you add:
>
> config MODULES
> boolean
> default y
>
> then you end up with CONFIG_A=m in the `.config'.
Yes, so it is the problem. I got CONFIG_A=m in .config from
CONFIG_A=y in defconfig.
> Now, I may have missed something.
>
> Do you have a situation where a MODULES symbol is present, enabled,
> but CONFIG_A=y still appear from CONFIG_A=m in the defconfig ?
I tested in this way:
1. Add this example to the end of drivers/char/Kconfig file.
-------------------------------------
choice
prompt "choice list AB"
config A
tristate "a"
config B
tristate "b"
endchoice
choice
prompt "choice list CD"
depends on A
config C
bool "c"
config D
bool "d"
endchoice
-------------------------------------
2. Then run commands:
$ cp arch/x86/configs/i386_defconfig arch/x86/configs/test_defconfig
$ echo 'CONFIG_A=y' >> arch/x86/configs/test_defconfig
$ make O=../build test_defconfig
$ grep CONFIG_A= ../build/.config
CONFIG_A=m
---
Atsushi Nemoto
On Thu, 09 Feb 2012 23:30:17 +0900 (JST), Atsushi Nemoto <[email protected]> wrote:
> I tested in this way:
>
> 1. Add this example to the end of drivers/char/Kconfig file.
> -------------------------------------
> choice
> prompt "choice list AB"
> config A
> tristate "a"
> config B
> tristate "b"
> endchoice
> choice
> prompt "choice list CD"
> depends on A
> config C
> bool "c"
> config D
> bool "d"
> endchoice
> -------------------------------------
>
> 2. Then run commands:
>
> $ cp arch/x86/configs/i386_defconfig arch/x86/configs/test_defconfig
> $ echo 'CONFIG_A=y' >> arch/x86/configs/test_defconfig
> $ make O=../build test_defconfig
> $ grep CONFIG_A= ../build/.config
> CONFIG_A=m
I did some investigations.
When this happens, sym_calc_value() was called recursively.
main
conf_set_all_new_symbols
sym_calc_value ("choice list CD")
sym_calc_visibility ("choice list CD")
expr_calc_value
sym_calc_value ("A")
On this place in sym_calc_value(), sym->def[S_DEF_USER].tri was 'yes'
and sym->visible was 'mod'.
newval.tri = EXPR_AND(sym->def[S_DEF_USER].tri,
sym->visible);
goto calc_newval;
Then new value of CONFIG_A will be 'mod'.
I suppose sym->visible is 'mod' because default value of tristate
choice is 'mod'. Is this correct?
Any suggestions for fixing or debugging?
---
Atsushi Nemoto