So nothing collides. A module called span collides with the half-dozen other things in an embedded codebase already called span, and with C++20's. spatium does not. The cost is The Latin, decoded; the benefit is never renaming anything again.
Because the footprint is then a number you can state before the program runs, in place of an observation about one run. The long answer, including what it costs you, is Why there is no malloc.
Yes, with no switch. A cellblock's boundaries, claims and counts are atomic words, each moved in one step, and every context that reaches a cellblock reaches it at once without waiting for another. A cell of either tier is given back by the one that holds it, in any order.
The one genuinely concurrent module is memoria_anularis, which is single-producer single-consumer only. Two producers on one ring is broken, and never merely slow. See Who owns what.
Nothing, given const. gcc devirtualizes a call through a static const <Mod>Ns to the inlined body, identical instructions to calling the entry directly, and does not devirtualize through a non-const one, which becomes a real call with an 88-byte frame. clang devirtualizes both. For that reason EMBED_TABLE_STORAGE includes const. See The dispatch table idiom.
Not for the hot entries: they are EMBED_INLINE in their headers. For cross-module calls that are not, it is worth a lot: mmgr_memor_chr over 512 bytes measured 610 cycles without and 187 with. MMGR_LTO defaults to ON.
Yes. Headers are wrapped in EMBED_BEGIN_DECLS / EMBED_END_DECLS, which is extern "C" under a C++ compiler. The dispatch tables are plain structs of function pointers and work unchanged.
Because -Os declines to inline the small branch-free entries that are this library, and the calls it saves space on are the calls that cost the most. The top-level CMakeLists.txt rewrites -Os to -O2 for that configuration. An embedded build reaching for MinSizeRel wants small, not slow-and-small.
There aren't five builds. One build produces five sets of targets. Each is the library at a different set of compile-time widths. A defect that only appears at a 16-bit carrier fails a run on your machine instead of waiting for the hardware. One cmake --build, one ctest. See The environment matrix.
No. memoriam_praetereo and memoria_externa are behind MMGR_ENABLE_DMA and MMGR_ENABLE_EXTRAM, both off by default, and their suites are skipped, loudly, with a CMake status message. Turn the flag on if you want them built. A capability gates the whole suite, never a case inside one, because a suite that compiles half its cases away still reports as passing.
Measure it, and turn the peak tracking on first. A cellblock's own counters and buf_available report where the tier boundaries are right now and never where they have been. Reading them after a workload tells you about that instant alone.
The peaks are persistent_hw and temporary_hw, one per tier, and they only exist and are only maintained when MMGR_ENABLE_HW_MEM_CAPACITY_CB is on. It is off by default: both stay zero and a reading from a default build means nothing. Build the checks environment with it on, run your real workload, read them, and size to the peak plus margin. Your first prison site has the procedure.
Because nothing was scrubbed on a minimum security pool. The memory is readable and holds whatever was there; the next take will hand it to someone else. A pointer that outlives its release is the sharpest edge in this library and nothing can detect it for you. Who owns what.
There is no bounded form of it: the target size is not one of its arguments, and there is nothing for the shim to map it onto. The name is simply not provided, and the shim also defines the <string.h> include guards so libc's declaration never arrives either. A call to it fails to compile instead of overflowing quietly. Use strlcpy, which takes the size. Dropping in for <string.h>.
Measured: the open-coded fold was 3.609 cycles against 3.680 for __builtin_ctzll, and __builtin_popcountll compiles to a libgcc call on baseline x86-64. They were not faster, and avoiding them keeps the scan path free of per-compiler #ifdefs. Doing eight bytes at a time.
The license text governs, not this page, and this is not legal advice. AGPL-3.0-or-later, in practice has the practical notes for a library consumed by add_subdirectory. If the terms do not suit your project, ask: dual licensing is settled in conversation, with no policy to read first.
