[tlbuild] lcdf-typetools-2.78 -- is hashcode() unsigned, uint32_t, or size_t

[Peter Breitenlohner]

On Thu, 30 Apr 2009, Angelo Graziosi wrote:

Hi Eddie,

here a report from an attempt to build lcdf-typetools for TeX Live.

>> On current Cygwin (1.5.25-15) it fails, without applying any *patches*, as 
>> follow:
>> CC='gcc-4' CXX='g++-4' CPP='cpp-4' CFLAGS='-Wno-write-strings
>> -Wno-attributes' CXXFLAGS='-Wno-write-strings -Wno-attributes'
>> LDFLAGS='-Wl,--enable-auto-import -Wl,--enable-runtime-pseudo-reloc'
>> TL_CONFIGURE_ARGS='--disable-xetex --without-graphite' ./Build
>> [...]
>> g++-4 -W -Wall -DHAVE_CONFIG_H -I. 
>> -I../../../../texk/lcdf-typetools/liblcdf -I.. 
>> -I../../../../texk/lcdf-typetools/include    -Wno-write-strings 
>> -Wno-attributes -MT string.o -MD -MP -MF .deps/string.Tpo -c -o string.o 
>> ../../../../texk/lcdf-typetools/liblcdf/string.cc
>> ../../../../texk/lcdf-typetools/liblcdf/string.cc:535: error: prototype for 
>> 'uint32_t String::hashcode(const char*, const char*)' does not match any in 
>> class 'String'
>> ../../../../texk/lcdf-typetools/include/lcdf/string.hh:261: error: 
>> candidates are: unsigned int String::hashcode() const
>> ../../../../texk/lcdf-typetools/include/lcdf/string.hh:250: error: 
>> static unsigned int String::hashcode(const unsigned char*, const unsigned 
>> char*)
>> ../../../../texk/lcdf-typetools/include/lcdf/string.hh:246: error: 
>> static unsigned int String::hashcode(const char*, const char*)
>> make[6]: *** [string.o] Error 1
> [...]
>
> Applying only this patch, seems to solve many problems...
> =======================================
> diff -Naur texlive-source.orig/texk/lcdf-typetools/include/lcdf/string.hh 
> texlive-source/texk/lcdf-typetools/include/lcdf/string.hh
> --- texlive-source.orig/texk/lcdf-typetools/include/lcdf/string.hh 2009-03-20 
> 10:50:07.000000000 +0100
> +++ texlive-source/texk/lcdf-typetools/include/lcdf/string.hh 2009-04-30 
> 10:16:19.328125000 +0200
> @@ -243,10 +243,10 @@
>      * @invariant If end1 - begin1 == end2 - begin2 and memcmp(begin1, 
> begin2,
>      * end1 - begin1) == 0, then hashcode(begin1, end1) == hashcode(begin2,
>      * end2). */
> -    static unsigned hashcode(const char *begin, const char *end);
> +    static uint32_t hashcode(const char *begin, const char *end);
>
>     /** @overload */
> -    static inline unsigned hashcode(const unsigned char *begin,
> +    static inline uint32_t hashcode(const unsigned char *begin,
>                                    const unsigned char *end) {
>        return hashcode(reinterpret_cast<const char *>(begin),
>                        reinterpret_cast<const char *>(end));
> @@ -258,7 +258,7 @@
>      * "SuperFastHash."
>      *
>      * @invariant  If s1 == s2, then s1.hashcode() == s2.hashcode(). */
> -    inline unsigned hashcode() const {
> +    inline uint32_t hashcode() const {
>        return length() ? hashcode(begin(), end()) : 0;
>     }
> =======================================

Interestingly this problem show up for neither Cygwin 1.7, Linux, Solaris,
nor Mac OS X.  The declaration, implementation, and use of hashcode() are,
however, inconsistent but I somewhat hesitate to apply to patch above
because the file include/lcdf/string.hh does not include any header that
might be required to declare uint32_t.

Can you please clarify this.

Regards
Peter