update unifdef causing partial header generation and

[Bernhard Reutner-Fischer]

On 7 January 2014 13:03, Vineet Gupta <Vineet.Gupta1 at synopsys.com> wrote:
> On Tuesday 07 January 2014 04:40 PM, Vineet Gupta wrote:
>> On Tuesday 07 January 2014 02:59 PM, Bernhard Reutner-Fischer wrote:
>>> On 3 January 2014 05:53, Vineet Gupta <Vineet.Gupta1 at synopsys.com> wrote:
>>>> On Monday 23 December 2013 07:46 PM, Bernhard Reutner-Fischer wrote:
>>>>
>> ------------------->8-----------------------------
>> ./extra/scripts/unifdef -B -t -f .//include/generated/unifdef_config.h -U_LIBC
>> -U__UCLIBC_GEN_LOCALE -U__NO_CTYPE include/bits/uClibc_config.h
>> #if !defined _FEATURES_H && !defined __need_uClibc_config_h
>> # error Never include <bits/uClibc_config.h> directly; use <features.h> instead
>> unifdef: include/bits/uClibc_config.h: 3: Inappropriate #endif
>> unifdef: output may be truncated
>>
>> I debugged it a bit and it seems removing -t option to unifdef seems to elide the
>> issue.
>> Attached are my symbol file and the src file - see if u can reproduce it at your end.
>
> Following fixes the issue - although it might be incomplete.
>
> -------------------->
> From 4f72593ad75f2c8a5a52501e020997f7d6e86aac Mon Sep 17 00:00:00 2001
> From: Vineet Gupta <vgupta at synopsys.com>
> Date: Tue, 7 Jan 2014 17:25:14 +0530
> Subject: [PATCH] Reset unifdef state machine after a -f <file> parsing
>
> After commit 2a021ae81c36 "buildsys: update unifdef" there were sporadic
> build failures at the time of uClibc header generation.
>
> ----------------->8------------------
> unifdef: include/bits/kernel_sigaction.h: 24: Inappropriate #endif
> unifdef: output may be truncated
> unifdef: include/bits/uClibc_config.h: 3: Inappropriate #endif
> ...
> ...
> ----------------->8------------------
>
> Turns out that unifdef now has -f <file> option to provide def/undef
> symbols from a file.
>
> This helper file parsing uses the same code as the SRC file parsing.
> However the parsing state machine uses global variables which need to
> be "reset" after the -f pass.
>
> Signed-off-by: Vineet Gupta <vgupta at synopsys.com>
> ---
>  extra/scripts/unifdef.c | 2 ++
>  1 file changed, 2 insertions(+)
>
> diff --git a/extra/scripts/unifdef.c b/extra/scripts/unifdef.c
> index b159df0a665b..f71ee66c896d 100644
> --- a/extra/scripts/unifdef.c
> +++ b/extra/scripts/unifdef.c
> @@ -378,6 +378,8 @@ processinout(const char *ifn, const char *ofn)
>  {
>         struct stat st;
>
> +       incomment = linestate = 0;
> +
>         if (ifn == NULL || strcmp(ifn, "-") == 0) {
>                 filename = "[stdin]";
>                 linefile = NULL;
> --
> 1.8.3.2

I have just pushed
http://git.uclibc.org/uClibc/commit/?id=c71f8bc18e33da575c2f637a4dfa5e6bf120cd3c

does that work for you, Vineet?

Tony, when parsing defundefile there is an error in the linestate
machine, it seems. For me, several entries are "swallowed", i.e. not
parsed unless i conditionalize the fgets in skiphash on linestate.
When the state machine is confused and leaves defundefile,
processinout then starts with a confused state and throws away lines
erroneously (which is what the hunk cited above would attempt to
workaround after the fact, i guess).

Please consider applying the 2 hunks in abovementioned commit c71f8bc
or let me know if that is either wrong or you need me to submit a
format patch.

TIA and cheers,