When repeatedly appending an object to a growing array, we would create a new
array larger by one slot, insert all the old entries and the new element with
ref count bumps into the new array, and then unref the old array.
This would cause problems when building an array with more than a few thousand
elements. If userdbctl is modified to construct an array,
'userdbctl --json=pretty group >/dev/null' with 31k groups:
0.74s (existing code)
102.17s (returning an array)
0.79s (with this patch)
We append arrays in various places, so it seems nice to make this generally
fast.
The source would be set implicitly when parsing from a named file. But
it's useful to specify the source also for cases where we're parsing a
ready string. I noticed the lack of this API when trying to write tests,
but it seems generally useful to be specify a source name when parsing
things.
Unfortunately, hex output can only be produced with unsigned types. Some
cases can be fixed by producing the correct type, but a few simply have
to be cast. At least casting makes it explicit.
This fixes formatting of JSON real values, and uses C locale for them.
It's kinda interesting that this wasn't noticed before: the C locale
object we allocated was not used, hence doing the dance had zero effect.
This makes "test-varlink" pass again on systems with non-C locale.
(My guess: noone noticed this because "long double" was used before by
the JSON code and that had no locale supporting printer or so?)
This normally wouldn't happen, but if some of those places were called
with lhs and rhs being the same object, we could unref the last ref first,
and then try to take the ref again. It's easier to be safe, and with the
helper we save some lines too.
The rest of our JSON code tries hard to magically convert NULL inputs
into "null" JSON objects, let's make sure this also works with
json_variant_set_field().
We were already asserting that the intmax_t and uintmax_t types
are the same as int64_t and uint64_t. Pretty much everywhere in
the code base we use the latter types. In principle intmax_t could
be something different on some new architecture, and then the code would
fail to compile or behave differently. We actually do not want the code
to behave differently on those architectures, because that'd break
interoperability. So let's just use int64_t/uint64_t since that's what
we indend to use.
It seems that the implementation of long double on ppc64el doesn't really work:
long double cast to integer and back compares as unequal to itself. Strangely,
this effect happens without optimization and both with gcc and clang, so it
seems to be an effect of how long double is implemented by the architecture.
Dumping the values shows the following pattern:
00 00 00 00 00 00 24 40 00 00 00 00 00 00 00 00 # long double v = 10;
00 00 00 00 00 00 24 40 00 00 00 00 00 00 80 39 # (long double)(intmax_t) v
Instead of trying to make this work, I think it's most reasonable to switch to
normal doubles. Notably, we had no tests for floating point behaviour. The
first test we added (for values even not in the range outside of double),
showed failures.
Common implementations of JSON (in particular JavaScript) use 64 bit double.
If we stick to this, users are likely to be happy when they exchange data with
those tools. Exporting values that cannot be represented in other tools would
just cause interop problems.
I don't think the extra precision would be much used. Long double seems to make
most sense as a transient format used in calculations to get extra precision in
operations, and not a storage or exchange format. So I expect low-level
numerical routines that have to know about hardware to make use of it, but it
shouldn't be used by our (higher-level) system library. In particular, we would
have to add tests for implementations conforming to IEEE 754, and those that
don't conform, and account for various implementation differences. It just
doesn't seem worth the effort.
https://en.wikipedia.org/wiki/Long_double#Implementations shows that the
situation is "complicated":
> On the x86 architecture, most C compilers implement long double as the 80-bit
> extended precision type supported by x86 hardware. An exception is Microsoft
> Visual C++ for x86, which makes long double a synonym for double. The Intel
> C++ compiler on Microsoft Windows supports extended precision, but requires
> the /Qlong‑double switch for long double to correspond to the hardware's
> extended precision format.
> Compilers may also use long double for the IEEE 754 quadruple-precision
> binary floating-point format (binary128). This is the case on HP-UX,
> Solaris/SPARC, MIPS with the 64-bit or n32 ABI, 64-bit ARM (AArch64) (on
> operating systems using the standard AAPCS calling conventions, such as
> Linux), and z/OS with FLOAT(IEEE). Most implementations are in software, but
> some processors have hardware support.
> On some PowerPC and SPARCv9 machines, long double is implemented as a
> double-double arithmetic, where a long double value is regarded as the exact
> sum of two double-precision values, giving at least a 106-bit precision; with
> such a format, the long double type does not conform to the IEEE
> floating-point standard. Otherwise, long double is simply a synonym for
> double (double precision), e.g. on 32-bit ARM, 64-bit ARM (AArch64) (on
> Windows and macOS) and on 32-bit MIPS (old ABI, a.k.a. o32).
> With the GNU C Compiler, long double is 80-bit extended precision on x86
> processors regardless of the physical storage used for the type (which can be
> either 96 or 128 bits). On some other architectures, long double can be
> double-double (e.g. on PowerPC) or 128-bit quadruple precision (e.g. on
> SPARC). As of gcc 4.3, a quadruple precision is also supported on x86, but as
> the nonstandard type __float128 rather than long double.
> Although the x86 architecture, and specifically the x87 floating-point
> instructions on x86, supports 80-bit extended-precision operations, it is
> possible to configure the processor to automatically round operations to
> double (or even single) precision. Conversely, in extended-precision mode,
> extended precision may be used for intermediate compiler-generated
> calculations even when the final results are stored at a lower precision
> (i.e. FLT_EVAL_METHOD == 2). With gcc on Linux, 80-bit extended precision is
> the default; on several BSD operating systems (FreeBSD and OpenBSD),
> double-precision mode is the default, and long double operations are
> effectively reduced to double precision. (NetBSD 7.0 and later, however,
> defaults to 80-bit extended precision). However, it is possible to override
> this within an individual program via the FLDCW "floating-point load
> control-word" instruction. On x86_64, the BSDs default to 80-bit extended
> precision. Microsoft Windows with Visual C++ also sets the processor in
> double-precision mode by default, but this can again be overridden within an
> individual program (e.g. by the _controlfp_s function in Visual C++). The
> Intel C++ Compiler for x86, on the other hand, enables extended-precision
> mode by default. On IA-32 OS X, long double is 80-bit extended precision.
So, in short, the only thing that can be said is that nothing can be said. In
common scenarios, we are getting only a bit of extra precision (80 bits instead
of 64), but use space for padding. In other scenarios we are getting no extra
precision. And the variance in implementations is a big issue: we can expect
strange differences in behaviour between architectures, systems, compiler
versions, compilation options, and even the other things that the program is
doing.
Fixes#21390.
JSON doesn't have NaN/infinity/-infinity concepts in the spec.
Implementations vary what they do with it. JSON5 + Python simply
generate special words "NAN" and "Inifinity" from it. Others generate
"null" for it.
At this point we never actually want to output this, so let's be
conservative and generate RFC compliant JSON, i.e. convert to null.
One day should JSON5 actually become a thing we can revisit this, but in
that case we should implement things via a flag, and only optinally
process nan/infinity/-infinity.
This patch is extremely simple: whenever accepting a
nan/infinity/-infinity from outside it converts it to NULL. I.e. we
convert on input, not output.
Compound statements is this stuff: ({ … })
Compound literals is this stuff: (type) { … }
We use compound statements a lot in macro definitions: they have one
drawback though: they define a code block of their own, hence if macro
invocations are nested within them that use compound literals their
lifetime is limited to the code block, which might be unexpected.
Thankfully, we can rework things from compound statements to compund
literals in the case of json.h: they don't open a new codeblack, and
hence do not suffer by the problem explained above.
The interesting thing about compound statements is that they also work
for simple types, not just for structs/unions/arrays. We can use this
here for a typechecked implicit conversion: we want to superficially
typecheck arguments to the json_build() varargs function, and we do that
by assigning the specified arguments to our compound literals, which
does the minimal amount of typechecks and ensures that types are
propagated on correctly.
We need one special tweak for this: sd_id128_t is not a simple type but
a union. Using compound literals for initialzing that would mean
specifiying the components of the union, not a complete sd_id128_t. Our
hack around that: instead of passing the object directly via the stack
we now take a pointer (and thus a simple type) instead.
Nice side-effect of all this: compound literals is C99, while compound
statements are a GCC extension, hence we move closer to standard C.
Fixes: #20501
Replaces: #20512
In general we almost never hit those asserts in production code, so users see
them very rarely, if ever. But either way, we just need something that users
can pass to the developers.
We have quite a few of those asserts, and some have fairly nice messages, but
many are like "WTF?" or "???" or "unexpected something". The error that is
printed includes the file location, and function name. In almost all functions
there's at most one assert, so the function name alone is enough to identify
the failure for a developer. So we don't get much extra from the message, and
we might just as well drop them.
Dropping them makes our code a tiny bit smaller, and most importantly, improves
development experience by making it easy to insert such an assert in the code
without thinking how to phrase the argument.
We recently started making more use of malloc_usable_size() and rely on
it (see the string_erase() story). Given that we don't really support
sytems where malloc_usable_size() cannot be trusted beyond statistics
anyway, let's go fully in and rework GREEDY_REALLOC() on top of it:
instead of passing around and maintaining the currenly allocated size
everywhere, let's just derive it automatically from
malloc_usable_size().
I am mostly after this for the simplicity this brings. It also brings
minor efficiency improvements I guess, but things become so much nicer
to look at if we can avoid these allocation size variables everywhere.
Note that the malloc_usable_size() man page says relying on it wasn't
"good programming practice", but I think it does this for reasons that
don't apply here: the greedy realloc logic specifically doesn't rely on
the returned extra size, beyond the fact that it is equal or larger than
what was requested.
(This commit was supposed to be a quick patch btw, but apparently we use
the greedy realloc stuff quite a bit across the codebase, so this ends
up touching *a*lot* of code.)
Prompted by https://bugzilla.redhat.com/show_bug.cgi?id=1930875 in which
I had previously used json_dispatch_unsigned and passed a return variable of
type unsigned when json_dispatch_unsigned writes a uintmax_t.
This adds a new flag JSON_FORMAT_OFF that is a marker for "no JSON
output please!".
Of course, this flag sounds pointless in a JSON implementation, however
this is useful in code that can generate JSON output, but also more
human friendly output (for example our table formatters).
With this in place various tools that so far maintained one boolean
field "arg_json" that controlled whether JSON output was requested at
all and another field "arg_json_format_flags" for selecing the precise
json output flags may merge them into one, simplifying code a bit.
This is similar to the base64 support, but fixed-size hash values are
typically preferably presented as series of hex values, hence store them
here like that too.
E.g. in nss-resolve it is still useful to print the location of the error:
src/test/test-nss.c:231: dlsym(0x0x1dc6fb0, _nss_resolve_gethostbyname2_r) → 0x0x7fdbfc53f626
(string):1:40: JSON field ifindex is out of bounds for an interface index.
I opted to use a partially duplicated if condition to avoid nesting. It's nice
to have the log calls vertically aligned. The compiler will optimize this nicely.
Yu Watanabe
Zbigniew Jędrzejewski-Szmek
Lennart Poettering
David Tardon
Jan Janssen
Mike Gilbert
Anita Zhang
rnhmjoj