Fixes warnings on versions of Linux where mallinfo(3C) has
been deprecated.

The glibc crypt() implementation in RHEL 8 apparently sometimes
performs misguided "validation" of the salt even for "$6$" and
fails with EINVAL instead of actually validating the password.

Throwing an error on validation failure is not what existing
code expects, so just return a failure instead in this case.

Fixes #10050.

See also https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=714219

Martin Nilsson authored 7 years ago and Marcus Comstedt committed 4 years ago

Fix compilation issue with f_gethrdtime without threads. Don't know if the result makes sense though.

Adds variant using timezone/altzone (eg Solaris).

Fixes the fallback implementation (using gmtime(3F)) to not
drift by an hour when the local time zone has tm_isdst == 1.

string_to_utf8() now supports the same extensions as utf8_to_string().

Fixes [PIKE-111].

It now seems to work as intended also when include_subclasses is true.

Debug.find_all_clones() is an efun, so it must follow the
strict pike stack calling conventions.

Got rid of args to mktime_zone error. Fixed use of uninitialized value.
Indent (pretend) class local function.
Removed unused set_zone.
struct tm requires time_stuff.
Got rid of (now unused) fname argument too.

System.TM: Properly fixed, code reuse with mktime().

Extended testsuite for both.
Reduced memory footprint.

It should now accept being called without arguments again.

Thanks to Jeff Hungerford <hungerf3-roxen3@house.ofdoom.com>
for the report.

Fixes [Pike mailinglist 14512]/[LysLysKOM 21651115].

It is now possible to request a quick gc of a mapping with weak references.

Adds an optional second argument to unicode_to_string() to
specify the default byte order (an initial BOM in the input
string overrides this argument).

Adds an optional second argument to string_to_unicode() to
specify the wanted byte order.

The fields in struct mallinfo in Linux libc (as of glibc 2.20) are
declared as int, and will thus overflow into negative when the values
get larger than 0x3fffffff. This commit interprets negative mallinfo
values as 32-bit unsigned values.

This change fixes two bugs
1) the overflow checks were incorrect, because signed integers do not
   overflow
2) when overflow was actually detected, the code fell back to the
   slow path (using f_add) with the last array element as the new start
   value, which led to wrong results

It is now way faster for mappings (in testing about 2x) and somewhat
faster for multisets. Since our developers have a tendency to write
if(has_index(map,key)) instead of if(mapping[key]) this can be a
noticeable gain.

One of them was actually legitimate.