[tz] source code changes in tzcode2013d: int_:fast32_t

Christos Zoulas christos at zoulas.com
Sun Jul 21 14:25:49 UTC 2013


On Jul 21,  1:20pm, alois at astro.ch (Alois Treindl) wrote:
-- Subject: Re: [tz] source code changes in tzcode2013d: int_:fast32_t

| I noticed that a lot of source code changes were done in tzcode, for 
| example many variables which were of type long have been changed
| to int_fast32_t.
| 
| On my system, a x86_64 Linux, we get: typedef long int_fast32_t.
| So as expected, it comes out the same. long is 64 bit.
| 
| But I wonder about the rationale behind the naming of this type: is it 
| supposed to hold only 32-bit values? If it is supposed not to suffer 
| from 32-bit overflow problems, i.e. meant to be wider than 32 bits, why 
| is it named the way it is named?
| 
| A general question: there is very little documentation for the source 
| code in general. Would it not be appropriate for the maintainers of the 
| code to include more internal documentation?

The biggest problem with this change is that it is ABI changing in the
following 3 functions:

struct tm *
offtime(const time_t * clock, long int offset);

struct tm *
offtime_r(const time_t * clock, long int offset, struct tm *ret);

time_t
timeoff(struct tm * tm, long int offset);

In order not to change the ABI, I did not change them for NetBSD.

The rest of the changes are internal, so they don't matter.

And a cosmetic one, all the gcc attributes (not just some) should
be __foo__ instead of foo, and the macros should be named consistenty.

Finally my shameless plug, the NetBSD sources also contain the following
changes:
	- multi-timezone support, thread safe code (_rz functions)
	- code to behave like glibc in the DST gap (not fail if enabled)
	- consistent errno setting on failure
	- man pages in mdoc not man format
	- modernization (s/sprintf/snprintf/, s/register//)

christos



More information about the tz mailing list