Dear Colleagues,
An ongoing thread in the gnu-prog-discuss mailing list is discussing
persistent memory for GNU projects. Circa late February, Jacob Bachmeyer
suggested on that list that Guile might benefit from persistent memory.
After consulting the Guile manual, I'm writing now to outline what this
might mean and to invite interested persons in the Guile community to
discuss further.
First, a note on terminology: By "persistent memory" I do *not* mean
newfangled non-volatile memory hardware, e.g., Intel Optane or NVDIMMs.
Instead, "persistent memory" here refers to a purely software abstraction
that can be implemented on conventional hardware --- ordinary computers
with volatile DRAM memory and conventional block storage (HDDs, SSDs).
The easiest way to vividly explain the possibilities for Guile might be to
point out the persistent memory feature recently added to the GNU AWK
(gawk) interpreter. The new persistent memory feature of gawk is called
"pm-gawk." When you run an AWK script in the gawk interpreter, you may
activate pm-gawk by specifying a file containing a persistent heap where
pm-gawk will hold AWK variables and functions. AWK scripts themselves are
*not* modified in any way to activate pm-gawk. A persistent heap lives in
a file on disk and hangs around until deleted. It may be re-used by
subsequent runs of the same AWK script or by completely different AWK
scripts, i.e., it's a way to pass variables and functions between
unrelated scripts. As a side effect, pm-gawk gives AWK (the moral
equivalent of) a REPL. The Persistent Memory Gawk User Manual explains
the details; see the Quick Start section for a succinct concrete
illustration of use:
https://www.gnu.org/software/gawk/manual/pm-gawk/
If Guile did something similar, the details might differ from pm-gawk, but
the benefits would hopefully be the same: Efficient, transparent, and
nearly effortless persistence-to-disk for arbitrary Scheme code, at the
cost of small and fairly easy changes to Guile. Implementing pm-gawk was
remarkably easy --- under 100 lines of code added/changed in the initial
proof-of-concept prototype, most of which was integrating a
malloc-compatible persistent memory library into the gawk interpreter:
https://dl.acm.org/doi/pdf/10.1145/3534855
Another precedent for persistence in Scheme is a project that used an
older persistent heap of mine to retrofit persistence onto a Scheme
interpreter:
https://github.com/tvcutsem/schemeken
If you think that a similar persistence capability might be useful for
Guile, let's talk.
Thank you.
Terence Kelly
https://dl.acm.org/profile/81100523747
