Now with a bit more granularity for possibly-future-optional on-disk
data structures:
LFSR_RCOMPAT_NONSTANDARD 0x0001 ---- ---- ---- ---1 (reserved)
LFSR_RCOMPAT_MLEAF 0x0002 ---- ---- ---- --1-
LFSR_RCOMPAT_MSHRUB 0x0004 ---- ---- ---- -1-- (reserved)
LFSR_RCOMPAT_MTREE 0x0008 ---- ---- ---- 1---
LFSR_RCOMPAT_BSPROUT 0x0010 ---- ---- ---1 ----
LFSR_RCOMPAT_BLEAF 0x0020 ---- ---- --1- ----
LFSR_RCOMPAT_BSHRUB 0x0040 ---- ---- -1-- ----
LFSR_RCOMPAT_BTREE 0x0080 ---- ---- 1--- ----
LFSR_RCOMPAT_GRM 0x0100 ---- ---1 ---- ----
LFSR_WCOMPAT_NONSTANDARD 0x0001 ---- ---- ---- ---1 (reserved)
LFSR_OCOMPAT_NONSTANDARD 0x0001 ---- ---- ---- ---1 (reserved)
This adds a couple reserved flags:
- LFSR_*COMPAT_NONSTANDARD - This flag will never be set by a standard
version of littlefs. The idea is to allow implementations with
non-standard extensions a way to signal potential compatibility issues
without worrying about future compat flag conflicts.
This is limited to a single bit, but hey, it's not like it's possible
to predict all future extensions.
If a non-standard extension needs more granularity, reservations of
standard compat flags can always be requested, even if they don't end
up implemented in standard littlefs. (Though such reservations will
need a strong motivation, it's not like these flags are free).
- LFSR_RCOMPAT_MSHRUB - In theory littlefs supports a shrubbed mtree,
where the root is inlined into the mroot. But in practice this turned
out to be more complicated than it was worth. Still, a future
implementation may find an mshrub useful, so preserving a compat flag
for such a case makes sense.
That being said, I have no plans to add support for mshrubs even in
the dbg scripts.
I would like the expected feature-set for debug tools to be
well-defined, but also conservative. This gets a bit tricky with
theoretical features like the mshrubs, but until mshrubs are actually
implemented in littlefs, I would like to consider them non-standard.
The implication of this is that, while LFSR_RCOMPAT_MSHRUB is
currently "reserved", it may be repurposed for some other meaning in
the future.
These changes also rename *COMPATFLAGS -> *COMPAT, and reorder the tags
by decreasing importance. This ordering seems more valuable than the
original intention of making rcompat/wcompat a single bit flip.
Implementation-wise, it's interesting to note the internal-only
LFSR_*COMPAT_OVERFLOW flag. This gets set when out-of-range bits are set
on-disk, and allows us to detect unrepresentable compat flags without
too much extra complexity.
The extra encoding/decoding overhead does add a bit of cost though:
code stack
before: 33944 2880
after: 34124 (+0.5%) 2880 (+0.0%)
Christopher Haster
130281ac05