On Tuesday, 19 November 2013 at 10:38:06 UTC, Jonathan M Davis
wrote:
It _is_ true however that if you're not careful about it, you
can end up with
a lot of slices that keep whole blocks of memory from being
collected when
they don't really need to refer to that memory anymore. So,
depending on what
profiling shows, some applications may need to make adjustments
to avoid having
slices keep too much extraneous memory from being collected.
So, it's
something to keep in mind, but I defintely don't think that we
should be
changing our approach at all.
- Jonathan M Davis
I ended up doing this in DCD. The lexing step sliced the source
code, so when caching autocompletion information for all of
Phobos, Druntime, etc, the memory usage would get fairly large.
Adding a few ".dup" calls in the phase that converts the AST to
the autocompletion cache structures greatly reduced the memory
usage.
