Optimizaion: instead of reading all non modified JS files
from the cache into the memory at once, the patch allows
copying non-modified JS files from disk one by one
during the compilation output writing routine.
Optimization: instead of calculating fingerprints manually each time
the patch allows using precalculated fingerprints from the klib.
Optimization: share one md5 digest for all IC infrastructure.
Optimization: often we don't need all hashes for all available symbols.
This patch allows the calculation of the symbol hash only on demand.
This patch also allows cycles for the inline function dependency
graph in the IC infrastructure. Note that the inline function cycles
may crash the inliner, however it is not the IC infrastructure
responsibility to check the cycles.
If inline function A calls another inline function B,
we must use the original version of inline function A for inlining,
which doesn’t have inlined function B. Because during the inlining
process, we remap all occurrences of inline function A
to a temporary copy of function A, and if the function B
somehow uses function A (e.g. callable reference),
the built IR will have a reference to the temporary function,
not the original one. All these things lead to broken cross-module references.
This patch saves the original versions of all inline functions
before inlining and provides them during the inline process.
^KT-55930 Fixed
We do not need to check a default implementation of the interface during
the translation to JS because it must be checked before.
Moreover, this check breaks the produced JS code
if IR is partial loaded, e.g. during the incremental rebuild.
^KT-55716 Fixed
The patch adds an error if the module can not find the cross-module reference.
The patch removes the DCE optimization which eliminates implement() intrinsic,
because it leads to a broken cross-module reference and
broken JS code with implement() call, albeit in an unreachable block.
Basically, some package names were Native-specific, whilst the packages
themselves were not Native-specific at all. This was already reflected
in the directory layout, but not in the package names.
This is fixed here.
NFC, just an automatic rename of packages with fixes of imports.
Generally, the library order doesn't matter.
However, after lowering we need to post process
lowered klibs in topological order:
- commit incremental cache artifacts,
- produce JS AST.
The patch uses an ability of the Kotlin library resolver to load klibs
in topological order and drops its reimplementation from IC infrastructure.
There is one side effect:
klibs which are not reachable from the main module, will produce their JS.
This should be more correct than just ignoring them.
Artem Kobzar