On Sat, Jul 29, 2023 at 12:23 AM Yoann Congal <yoann.congal@xxxxxxxx> wrote: > > > Hi, > > While analyzing a Yocto bug[0] I think I've identified a problem in kconfig. > The problem happens if you have a hex or int type config without a default value. > Like this : > config TEST_KCONFIG > hex "Test kconfig" > # No default value > ... and try to start oldconfig with a closed stdin (like we have in Yocto): > echo -n "" | make oldconfig > > When this happens, oldconfig prompts for the value of TEST_KCONFIG but stdin is closed it get the global default value : an empty string. This is not a valid hex/int value so it prompts again, hence the infinite loop. > > I'm having trouble pointing where the bug is exactly : > * Should the global default value for hex/int be valid in their context? (like the minimal value of the range or 0/0x0) > * Must all int/hex config provide a valid default value? (This is the case for hex config in the kernel). This would have to be documented somewhere (Some other KConfig implementation did [1]) Presumably, it is reasonable to require explicit 'default' for int/hex. Most of the int/hex entries in Linux are already doing it. > * Should all oldconfig/syncconfig/... exit with an error when trying to prompt on a closed stdin? (I might be able to send a patch for this one) No. I have seen multiple scripts piping the 'yes' command to Kconfig. There is no reason to prohibit pipe or redirection. I think the fastest fix is to send a patch to U-Boot to add a default for CONFIG_DEBUG_UART_BASE (and more patches if there are other similar cases). > > NB: I know of olddefconfig but in the case of U-boot, syncconfig is used at a later stage and the problem appears again. > > What do you think? > > Regards, > > [0]: https://bugzilla.yoctoproject.org/show_bug.cgi?id=14136 > [1]: https://docs.zephyrproject.org/1.14.0/guides/kconfig/index.html#redundant-defaults > -- > Yoann Congal > Smile ECS - Tech Expert -- Best Regards Masahiro Yamada
