Message-ID: <CAFkk2KTStSumsC1ZESgttJCHw6xJXA5PL3-cUY81AE9piOnmRw@mail.gmail.com>
Date:   Wed, 18 Apr 2018 10:33:06 +0200
From:   Ulf Magnusson <ulfalizer@...il.com>
To:     Masahiro Yamada <yamada.masahiro@...ionext.com>
Cc:     Linux Kbuild mailing list <linux-kbuild@...r.kernel.org>,
        Linus Torvalds <torvalds@...ux-foundation.org>,
        Sam Ravnborg <sam@...nborg.org>,
        Nicholas Piggin <npiggin@...il.com>,
        Kees Cook <keescook@...omium.org>,
        Emese Revfy <re.emese@...il.com>, X86 ML <x86@...nel.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH 17/30] Documentation: kconfig: document a new Kconfig
 macro language

On Tue, Apr 17, 2018 at 5:07 PM, Masahiro Yamada
<yamada.masahiro@...ionext.com> wrote:
> 2018-04-15 17:08 GMT+09:00 Ulf Magnusson <ulfalizer@...il.com>:
>> On Fri, Apr 13, 2018 at 7:06 AM, Masahiro Yamada
>> <yamada.masahiro@...ionext.com> wrote:
>>> Add a document for the macro language introduced to Kconfig.
>>>
>>> The motivation of this work is to move the compiler option tests to
>>> Kconfig from Makefile.  A number of kernel features require the
>>> compiler support.  Enabling such features blindly in Kconfig ends up
>>> with a lot of nasty build-time testing in Makefiles.  If a chosen
>>> feature turns out unsupported by the compiler, what the build system
>>> can do is either to disable it (silently!) or to forcibly break the
>>> build, despite Kconfig has let the user to enable it.
>>>
>>> This change was strongly prompted by Linus Torvalds.  You can find
>>> his suggestions [1] [2] in ML.  The original idea was to add a new
>>> 'option', but I found generalized text expansion would make Kconfig
>>> more powerful and lovely.  While polishing up the implementation, I
>>> noticed sort of similarity between Make and Kconfig.  This might be
>>> too immature to be called 'language', but anyway here it is.  All
>>> ideas are from Make (you can even say it is addicted), so people
>>> will easily understand how it works.
>>>
>>> [1]: https://lkml.org/lkml/2016/12/9/577
>>> [2]: https://lkml.org/lkml/2018/2/7/527
>>>
>>> Signed-off-by: Masahiro Yamada <yamada.masahiro@...ionext.com>
>>> ---
>>>
>>> Changes in v3: None
>>> Changes in v2: None
>>>
>>>  Documentation/kbuild/kconfig-macro-language.txt | 179 ++++++++++++++++++++++++
>>>  MAINTAINERS                                     |   2 +-
>>>  2 files changed, 180 insertions(+), 1 deletion(-)
>>>  create mode 100644 Documentation/kbuild/kconfig-macro-language.txt
>>>
>>> diff --git a/Documentation/kbuild/kconfig-macro-language.txt b/Documentation/kbuild/kconfig-macro-language.txt
>>> new file mode 100644
>>> index 0000000..1f6281b
>>> --- /dev/null
>>> +++ b/Documentation/kbuild/kconfig-macro-language.txt
>>> @@ -0,0 +1,179 @@
>>> +Concept
>>> +-------
>>> +
>>> +The basic idea was inspired by Make. When we look at Make, we notice sort of
>>> +two languages in one. One language describes dependency graphs consisting of
>>> +targets and prerequisites. The other is a macro language for performing textual
>>> +substitution.
>>> +
>>> +There is clear distinction between the two language stages. For example, you
>>> +can write a makefile like follows:
>>> +
>>> +    APP := foo
>>> +    SRC := foo.c
>>> +    CC := gcc
>>> +
>>> +    $(APP): $(SRC)
>>> +            $(CC) -o $(APP) $(SRC)
>>> +
>>> +The macro language replaces the variable references with their expanded form,
>>> +and handles as if the source file were input like follows:
>>> +
>>> +    foo: foo.c
>>> +            gcc -o foo foo.c
>>> +
>>> +Then, Make analyzes the dependency graph and determines the targets to be
>>> +updated.
>>> +
>>> +The idea is quite similar in Kconfig - it is possible to describe a Kconfig
>>> +file like this:
>>> +
>>> +    CC := gcc
>>> +
>>> +    config CC_HAS_FOO
>>> +            def_bool $(shell $(srctree)/scripts/gcc-check-foo.sh $(CC))
>>> +
>>> +The macro language in Kconfig processes the source file into the following
>>> +intermediate:
>>> +
>>> +    config CC_HAS_FOO
>>> +            def_bool y
>>> +
>>> +Then, Kconfig moves onto the evaluation stage to resolve inter-symbol
>>> +dependency, which is explained in kconfig-language.txt.
>>> +
>>> +
>>> +Variables
>>> +---------
>>> +
>>> +Like in Make, a variable in Kconfig works as a macro variable.  A macro
>>> +variable is expanded "in place" to yield a text string that may then expanded
>>> +further. To get the value of a variable, enclose the variable name in $( ).
>>> +As a special case, single-letter variable names can omit the parentheses and is
>>> +simply referenced like $X. Unlike Make, Kconfig does not support curly braces
>>> +as in ${CC}.
>>
>> Do we need single-letter variable names for anything? It looks like
>> we're deviating
>> a bit from Make behavior already.
>>
>> I suspect they're just a side effect of Make having automatic variables like $@.
>> The Make manual discourages them otherwise:
>>
>> "A dollar sign followed by a character other than a dollar sign,
>> open-parenthesis or
>> open-brace treats that single character as the variable name. Thus, you could
>> reference the variable x with `$x'. However, this practice is strongly
>> discouraged,
>> except in the case of the automatic variables (see section Automatic
>> Variables)."
>>
>
> OK.  We do not need two ways to do the same thing.
>
> I will consider it
> although supporting single-letter variable is not costly.
>
>
>
> --
> Best Regards
> Masahiro Yamada

Can you think of any cases where dynamic generation of Kconfig symbol
names would be a good solution by the way?

Personally, I'd make the semantics obvious by requiring quotes, so you
could do e.g.

    bar := "bar value"
    baz := "baz value"
    foo := " $(bar)  $(baz) "

and get " bar value  baz value " in 'foo'.

That'd get rid of those $(empty) hacks and the like, and would make
things super easy to understand.

If expansion was then also limited to within "" (so that you always
produce values), things would get absolutely ridiculously simple, both
behavior- and implementation-wise. All the unput() shenanigans would
disappear (shaving off code), and everyone would be able to tell at a
glance how stuff works.

I know that you're trying to create something that will be immediately
familiar to people used to Make, and that allows things to be easily
moved over, but I wonder if it's worthwhile to carry over the parts that
people find the most confusing and subtle about Make. Simply and
recursively expanded variables and functions are great and make sense,
but the rest I'm not so sure about, especially in Kconfig context.

When I see $(foo) outside of quotes, I assume a textual substitution,
like in Make, because it sticks out like a sore thumb compared to the
rest of the Kconfig syntax (I also think "Perl scalar"). It's actually a
token substitution here. "$(foo)" makes it obvious.

Cheers,
Ulf

Powered by blists - more mailing lists