[PATCH 1/6] generated character data for libc/ctype

[Thomas Wolff]

On 13.03.2018 13:49, Corinna Vinschen wrote:
> On Mar 13 12:02, Can Finner wrote:
>> On Tue, Mar 13, 2018 at 11:21 AM, Can Finner <can.finner@gmail.com> wrote:
>>> On Tue, Mar 13, 2018 at 10:40 AM, Corinna Vinschen <vinschen@redhat.com> wrote:
>>>> On Mar 13 10:33, Can Finner wrote:
>>>>> On Mon, Mar 12, 2018 at 10:42 AM, Corinna Vinschen <vinschen@redhat.com> wrote:
>>>>>> On Mar  8 00:18, Thomas Wolff wrote:
>>>>>>> Makefile add-ons for both patch series (libc/string and libc/ctype) will be
>>>>>>> sent separately.
>>>>>>>  From 4cd871bea1c6cf677d57587a7e844bb9cb3b19be Mon Sep 17 00:00:00 2001
>>>>>>> From: Thomas Wolff <towo@towo.net>
>>>>>>> Date: Sun, 25 Feb 2018 16:29:33 +0100
>>>>>>> Subject: [PATCH 1/6] generated case conversion data, Unicode 10.0
>>>>>>> [...]
>>>>>> Patchset pushed.  I just squashed the makefile patch into the previous
>>>>>> patch to avoid a repository state not building.
>>>>> Hi,
>>>>> This patch breaks arm-none-eabi cross toolchain build with below error message:
>>>>>
>>>>> /data/.../obj/gcc1/gcc/xgcc -B/data/.../obj/gcc1/gcc/
>>>>> -B/data/.../obj/newlib/arm-none-eabi/thumb/newlib/ -isystem
>>>>> /data/.../obj/newlib/arm-none-eabi/thumb/newlib/targ-include -isystem
>>>>> /data/.../newlib-cygwin/newlib/libc/include
>>>>> -B/data/.../obj/newlib/arm-none-eabi/thumb/libgloss/arm
>>>>> -L/data/.../obj/newlib/arm-none-eabi/thumb/libgloss/libnosys
>>>>> -L/data/.../newlib-cygwin/libgloss/arm  -mthumb
>>>>> -DPACKAGE_NAME=\"newlib\" -DPACKAGE_TARNAME=\"newlib\"
>>>>> -DPACKAGE_VERSION=\"3.0.0\" -DPACKAGE_STRING=\"newlib\ 3.0.0\"
>>>>> -DPACKAGE_BUGREPORT=\"\" -DPACKAGE_URL=\"\" -I.
>>>>> -I/data/.../newlib-cygwin/newlib/libc/ctype -D__NO_SYSCALLS__
>>>>> -D_COMPILING_NEWLIB -fno-builtin      -g -ffunction-sections
>>>>> -fdata-sections -O2  -mthumb -c -o lib_a-categories.o `test -f
>>>>> 'categories.c' || echo
>>>>> '/data/.../newlib-cygwin/newlib/libc/ctype/'`categories.c
>>>>> /data/.../newlib-cygwin/newlib/libc/ctype/categories.c:5:17: error:
>>>>> width of 'cat' exceeds its type
>>>>>     enum category cat: 11;
>>>>>                   ^~~
>>>> I don't understand this error.  Why is an enum < 11 bits?!?
>>> To be honest, I don't either.  Will collect more information about this.
>> So it looks like arm ABI requires -fshort-enums for bare-metal
>> toolchain, which will pack enum type into small int-type.
>> In this case, the enum category as below has fewer than 256 entries in
>> call cases?
>> enum category {
>> #include "categories.cat"
>> };
> Have a look at the file, it has barely 32 categories, so even a :5
> would suffice.
>
> Thomas, what was the idea here?  11 + 21 = 32, so was it just to
> fill the struct?
Yes, and to keep the Unicode value right-aligned. But 8/24 will do 
alike, please change it so.
Thomas

>    If so, we may want to redefine the struct, as
> suggested by Can:
>
>> If so, can we change it into below?  If not, is there any macro can be
>> used to differentiate situations?
>> struct _category {
>>    enum category cat: 8;
>>    unsigned int first: 24;
>>    unsigned short delta;
>> } __attribute__((packed));
> Thanks,
> Corinna
>