SML/NJ 110.99.6.1 library upgrade causes Ur/Web miscompilation #618
Description
Compiling @achlipala’s Ur/Web with MLton 20241230 produces an urweb binary that infinite-loops while compiling a trivial program:
$ cat trial.ur
val main : transaction page = return <xml><body>hi</body></xml>
$ cat trial.urp
$/string
trial
$ urweb -version
The Ur/Web compiler, version 20200209 + 55a881ff9b50d9e5c3b2fd564f5cd44a5cc5e6bc
$ urweb -verbose trial
Starting parseJob....
Finished parseJob.
Starting parse....
Finished parse.
Starting elaborate....
CHECK: Folder
CHECK: String
CHECK: Trial
Finished elaborate.
Starting unnest....
Finished unnest.
Starting explify....
Finished explify.
Starting corify....
Finished corify.
Starting core_untangle....
[never terminates]The $/string load in trial.urp is important; without it, urweb quickly and correctly compiles the program.
MLton 20210107 works fine, so I bisected between the two releases. git-bisect reports 88b75e5 as the culprit:
88b75e5ee35292077f8c549cb3c210e23a47d771 is the first bad commit
commit 88b75e5ee35292077f8c549cb3c210e23a47d771
Author: Matthew Fluet <matthew.fluet@gmail.com>
Date: Sat Dec 14 15:29:31 2024 -0500
Update to SML/NJ 110.99.6.1 libraries
CHANGELOG.adoc | 3 +
doc/guide/src/CKitLibrary.adoc | 4 +-
doc/guide/src/MLLPTLibrary.adoc | 4 +-
doc/guide/src/MLRISCLibrary.adoc | 4 +-
doc/guide/src/SMLNJLibrary.adoc | 11 +-
lib/ckit-lib/ckit.tgz | Bin 200785 -> 200761 bytes
lib/mllpt-lib/ml-lpt.tgz | Bin 267067 -> 267376 bytes
lib/mlrisc-lib/MLRISC.tgz | Bin 1457337 -> 1457420 bytes
lib/smlnj-lib/smlnj-lib.patch | 2676 +++++++++++++++++++++++++-------------
lib/smlnj-lib/smlnj-lib.tgz | Bin 641512 -> 666400 bytes
10 files changed, 1820 insertions(+), 882 deletions(-)
bisect found first bad commit
I haven’t dug any further into this, but I wanted to go ahead and report it here in case there’s something in MLton’s library patch that is not playing well with Ur/Web. @MatthewFluet, how is lib/smlnj-lib/smlnj-lib.patch generated?
(I have not yet tried compiling Ur/Web with SML/NJ to see if non-termination manifests there too; that’s probably the right next step.)