bug#68244: hash-table improvements

[Andy Moreton]

On Mon 15 Jan 2024, Eli Zaretskii wrote:

>> From: Andy Moreton <andrewjmoreton@gmail.com>
>> Date: Sun, 14 Jan 2024 22:08:14 +0000
>> 
>> On Sat 13 Jan 2024, Mattias Engdegård wrote:
>> 
>> > All right, I've pushed the parts that I have little doubt about being a
>> > definite improvement: a batch of refactoring steps, then the hash-table
>> > printing reforms, the representation and growth algorithm improvements, and
>> > finally documentation updates, all now on master.
>> 
>> Something in this patch series causes a build failure on Windows using
>> the Mingw64 toolchain. I've tried bisecting this, and at each step doing
>> "git clean -xdf" and then autogen, configure and make for a native build.
>> 
>> The last good commit on master is:
>>   c3d0cc50faf5 ("Remove rehash-threshold and rehash-size struct members")
>> 
>> Bisect identifies this as the bad commit:
>>   d3cefd3e9835 ("Leaner hash table dumping and thawing")
>> 
>> I have included the end of the log from the failing build below.
>> 
>>     AndyM
>> 
>>                            -----o0o-----
>> 
>> cp -f temacs.exe bootstrap-emacs.exe
>> rm -f bootstrap-emacs.pdmp
>> ./temacs --batch  -l loadup --temacs=pbootstrap \
>>      --bin-dest c:/emacs/30.0.50/mingw64-x86_64-O2-native/bin/ --eln-dest 
>> c:/emacs/30.0.50/mingw64-x86_64-O2-native/lib/emacs/30.0.50/
>> Loading loadup.el (source)...
>> Dump mode: pbootstrap
>> Using load-path (c:/emacs/git/emacs/master/lisp 
>> c:/emacs/git/emacs/master/lisp/emacs-lisp 
>> c:/emacs/git/emacs/master/lisp/progmodes 
>> c:/emacs/git/emacs/master/lisp/language 
>> c:/emacs/git/emacs/master/lisp/international 
>> c:/emacs/git/emacs/master/lisp/textmodes c:/emacs/git/emacs/master/lisp/vc)
>> Loading emacs-lisp/debug-early (source)...
>> 
>> ...[snipped]...
>> 
>> Loading c:/emacs/git/emacs/master/lisp/emacs-lisp/rmc.el (source)...
>> Finding pointers to doc strings...
>> Finding pointers to doc strings...done
>> Dumping under the name bootstrap-emacs.pdmp
>> Dumping fingerprint: 
>> 9f59e49a3ba95f3d7451327de9cf2863b9ddb392220c5f88f384f16245a619c2
>> Dump complete
>> Byte counts: header=100 hot=16147196 discardable=207320 cold=11360080
>> Reloc counts: hot=1123689 discardable=5662
>> ANCIENT=yes make -C ../lisp compile-first EMACS="../src/bootstrap-emacs.exe"
>> make[3]: Entering directory 
>> '/c/emacs/git/emacs/master/build/mingw64-x86_64-O2-native/lisp'
>>   ELC      /c/emacs/git/emacs/master/lisp/emacs-lisp/macroexp.elc
>> 
>> C:/emacs/git/emacs/master/src/lisp.h:2564: Emacs fatal error: assertion 
>> failed: 0 < size
>
> Thanks, but I see no assertion 0 < size on line 2564 of lisp.h, or
> anywhere else in lisp.h.  Can you tell which assertion fails?

In commit d3cefd3e9835, line 2564 of lisp.h has "eassume (0 < size);" in
HASH_TABLE_SIZE().

If the hash table changes now allow zero sized tables then that would
trigger in a build with ENABLE_CHECKING. My build script had checking
enabled, which would explain the failure. Sorry for not mentioning that
earlier (I had forgotten that it was enabled).

With ENABLE_CHECKING disabled the build succeeds. Thus the invariants
need looking at to allow a checked build to work.

    AndyM