Re: Question on GCC 4.6 and -fpermissive

[Jonathan Wakely <jwakely dot gcc at gmail dot com>]

On 10 August 2011 12:02, Jeffrey Walton wrote:
> On Wed, Aug 10, 2011 at 5:32 AM, Jonathan Wakely <jwakely.gcc@gmail.com> wrote:
>> On 10 August 2011 09:22, Jeffrey Walton wrote:
>>> Hi Guys,
>>>
>>> I'm running a table drive test suite. The tables are about 250K each,
>>> and there are 7 of them. Each row in the table looks similar to the
>>> following (this particular row is consumed by a 'short'):
>>>
>>> ? ?{ 0x0001, 0x0001, <some bool> };
>>>
>>> Without '-fpermissive' the code would not compile'.
>>
>> Why? ?You might want to fix that. ?-fpermissive is a band-aid to work
>> around broken code, for backwards compatibility. ?You should aim to
>> wean yourself off it. You shouldn't write new code that relies on it.
> Agreed. Its old code, and the thought of performing casts on values in
> tables with files as large as 250K takes the wind out of me.
>
> Fortunately, its code from the test cases, and not used in production.
>
>>> With
>>> '-fpermissive', the code compiles but produces a warning for each line
>>> encountered:
>>>
>>> ? ?AdditionVerify.cpp:68:1: warning: narrowing conversion of ‘65535’
>>> ? ? ? ?from ‘int’ to ‘short int’ inside { } [-fpermissive]
>>>
>>> GCC's 4.6 docs do not appear to have a [no]warning switch for
>>> permissive [1] (the docs for the switches don't even mention the word
>>> 'permissive').
>>
>> It's listed in the options summary:
>> http://gcc.gnu.org/onlinedocs/gcc/Option-Summary.html
>> and here:
>> http://gcc.gnu.org/onlinedocs/gcc/C_002b_002b-Dialect-Options.html
> Right. I found the flag under the 4.6.1 dialect docs.
>
>> It's not a warning option, it alters the C++ language that's
>> supported, to allow non-standard constructs and things that were
>> "traditionally" allowed (either by pre-standard C++ compilers or by
>> earlier versions of G++)
>>
>> If you don't want -fpermissive then don't use it.
> I think I prefer -fpermissive in this particular case. Its going to
> take a while to track down the 1700 or so offensive values.
>
>>> Taking a stab in the dark, my current CXXFLAGS looks is
>>> below, which has not helped.
>>>
>>> ? ?CXXFLAGS += -std=c++0x -fpermissive -Wno-permissive \
>>> ? ? ? ?-Wno-narrow -Wno-narrowing -Wno-narrow-conversion
>>> ? ? ? ?-Wno-narrowing-conversion -Wno-coversion
>>
>> Well no, clearly just making up options isn't going to help.
> :) Fortunately, GCC consumes unknown options.
>
> In the absence of [clear] documentation that states what to do, I try
> to use creativity while searching for the answer.
>
>>> Any ideas on how to hush the compiler for for this warning. I'm
>>> concerned it the volume of output might be masking more relevant
>>> warnings.
>>
>> If you don't want warnings about narrowing then use -Wno-narrow, which
>> works for me.
> OK. ?-Wno-narrow did not [completely?] work for me on Fedora 15 and
> GCC 4.6. I say completely because there are 1734 warnings from GCC.
>
> Driver program:
> http://code.google.com/p/owasp-esapi-cplusplus/source/browse/trunk/deps/safeint/tests/TestMain.cpp
> Sample table: http://code.google.com/p/owasp-esapi-cplusplus/source/browse/trunk/deps/safeint/tests/AddVerify.cpp
>
> The makefile for the test program is located in the same folder
> (deps/safeint/tests) and not root:
> http://code.google.com/p/owasp-esapi-cplusplus/source/browse/trunk/deps/safeint/tests/GNUmakefile
>
>> If it doesn't work for you please provide a *working*
>> example, not pseudocode or one that doesn't actually demonstrate your
>> point.
> Agreed. The last example was poor. My apologies.
>
> Jeff
>
> [jeffrey@fedora15 ~]$ gcc --version
> gcc (GCC) 4.6.0 20110603 (Red Hat 4.6.0-10)
> Copyright (C) 2011 Free Software Foundation, Inc.
> This is free software; see the source for copying conditions. ?There is NO
> warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
>
> [jeffrey@fedora15 tests]$ make 2> make.txt
> g++ -DNDEBUG=1 -g -O2 -DSAFEINT_DISALLOW_UNSIGNED_NEGATION=1 -pipe
> -Wall -Wextra -Wno-type-limits -Wno-unused -std=c++0x -fpermissive
> -Wno-narrow -I../. TestMain.cpp IncDecVerify.cpp AddVerify.cpp
> SubVerify.cpp MultVerify.cpp DivVerify.cpp ModVerify.cpp
> UnaryVerify.cpp PtrVerify.cpp -o TestMain.exe
> ./TestMain.exe
> ...
> [jeffrey@fedora15 tests]$ cat make.txt | head -1
> IncDecVerify.cpp:427:1: warning: narrowing conversion of ‘128’ from
> ‘int’ to ‘char’ inside { } [-fpermissive]
> [jeffrey@fedora15 tests]$ cat make.txt | wc -l
> 1734

The combination of -std=c++0x and -fpermissive is an odd one.  "I want
to us C++0x features, but I want to accept crufty old pre-C++98 code".
 That's probably always going to cause some issues, and it isn't
likely to be a priority for anyone to do anything about that.