From makarius at sketis.net  Tue May  9 23:50:42 2023
From: makarius at sketis.net (Makarius)
Date: Tue, 9 May 2023 23:50:42 +0200
Subject: [isabelle-dev] Update to jdk-17.0.7
Message-ID: <5d37dc20-c903-5bf6-0923-9258ae982e08@sketis.net>

With Isabelle/896e255d4fc4 we are on jdk-17.0.7. This is a routine update from 
the OpenJDK project with a lot of minor changes, see also 
https://docs.azul.com/core/zulu-openjdk/release-notes/april-2023

Usually such OpenJDK updates are rock-solid and no problems to be expected.


	Makarius

From martin.desharnais at posteo.de  Sun May 28 23:00:33 2023
From: martin.desharnais at posteo.de (Martin Desharnais)
Date: Sun, 28 May 2023 21:00:33 +0000
Subject: [isabelle-dev] Proposal to add a new \<notapprox> symbol and
 corresponding glyph to the Isabelle font
Message-ID: <3c77b04e-fbc2-b6ba-fd50-1b5ef0119eca@posteo.de>

Dear Makarius, dear Isabelle developers,

the Isabelle font contains a glyph for the unicode character "≈" (U+2248 
Almost Equal To) usable through the symbol \<approx>.

I propose to add a glyph for its negation "≉" (U+2249 Not Almost Equal 
To). For the corresponding symbol name, I propose \<notapprox> similarly 
to how we have \<noteq> and \<notin>.

Note that I cannot write ¬(x≈y) instead, as my use case is to define 
abbreviations for positive and negative literals in a formalization of 
first-order logic with equality.

datatype 'a literal = Pos 'a | Neg 'a

abbreviation pos_eq (infix "≈" 50) where
   "t1 ≈ t2 ≡ Pos (Upair t1 t2)"

abbreviation neg_eq (infix "≉" 50) where
   "t1 ≉ t2 ≡ Neg (Upair t1 t2)"

Regards,
Martin Desharnais

From lp15 at cam.ac.uk  Mon May 29 13:21:01 2023
From: lp15 at cam.ac.uk (Lawrence Paulson)
Date: Mon, 29 May 2023 11:21:01 +0000
Subject: [isabelle-dev] Proposal to add a new \<notapprox> symbol and
 corresponding glyph to the Isabelle font
In-Reply-To: <3c77b04e-fbc2-b6ba-fd50-1b5ef0119eca@posteo.de>
References: <3c77b04e-fbc2-b6ba-fd50-1b5ef0119eca@posteo.de>
Message-ID: <668EEA86-035E-4CA5-BD9A-6EA42AC67B92@cam.ac.uk>

Looks reasonable to me!

Larry

> On 28 May 2023, at 22:00, Martin Desharnais <martin.desharnais at posteo.de> wrote:
> 
> I propose to add a glyph for its negation "≉" (U+2249 Not Almost Equal To). For the corresponding symbol name, I propose \<notapprox> similarly to how we have \<noteq> and \<notin>.
> 
> Note that I cannot write ¬(x≈y) instead, as my use case is to define abbreviations for positive and negative literals in a formalization of first-order logic with equality.


From makarius at sketis.net  Tue May 30 12:14:34 2023
From: makarius at sketis.net (Makarius)
Date: Tue, 30 May 2023 12:14:34 +0200
Subject: [isabelle-dev] NEWS: significantly reduced ML heap usage
Message-ID: <8ce946ef-f740-2aa3-fddb-87988e5c9a21@sketis.net>

*** General ***

* ML heap usage and stored heap size has been significantly reduced,
especially for applications with a lot of definitions in a 'locale' or
'class' context. The shrink factor is usually in the range 1.1 .. 2.0,
but a factor 3 .. 10 has been seen in unusual situations. This often
allows big applications to return to the "small" 64_32 memory model with
its hard limit of 16 GiB, and thus reduce the heap size by another
factor 1.8.


This refers to current f3d19c8445ec: many changesets are leading towards this 
great improvement. The key point is actually rather small:

changeset:   78048:f16067da45ef
user:        wenzelm
date:        Mon May 15 14:13:58 2023 +0200
files:       src/Pure/Isar/proof_context.ML src/Pure/assumption.ML 
src/Pure/variable.ML
description:
avoid capture of inner/outer context and thus reduce heaps sizes by 20..40% 
(see also dd04a8b654fc, e49bf4ebf330, 9c19e15c8548, 71467e35fc3c);


That factor 1.2 .. 1.4 was merely for HOL and HOL-Analysis. Better factors are 
seen here:

   HOL-Algebra  3.6
   HOL-Probability 1.6

   Category3  5.2
   MonoidalCategory   7.6


	Makarius

From makarius at sketis.net  Tue May 30 22:10:23 2023
From: makarius at sketis.net (Makarius)
Date: Tue, 30 May 2023 22:10:23 +0200
Subject: [isabelle-dev] NEWS: significantly reduced ML heap usage
In-Reply-To: <8ce946ef-f740-2aa3-fddb-87988e5c9a21@sketis.net>
References: <8ce946ef-f740-2aa3-fddb-87988e5c9a21@sketis.net>
Message-ID: <60de0191-7c63-973c-b907-0168d6e74669@sketis.net>

Here are some more measurements to show the quite significant effect on 
locale-laden applications.

This is (1) current Isabelle/f3d19c8445ec + AFP/d34609a6a678 versus (2) 
Isabelle2022 + afp-2022/2458cc9f2178 on the rather heavy session Padic_Ints.

The whole stack of contributing sessions is shown below, with shrink factor:

HOL 1.56
HOL-Library 1.43
HOL-Computational_Algebra 1.80
HOL-Algebra 4.03
Padic_Ints 5.18
Padic_Field 12.5
ALL 2.87 (1.45 GiB / 0.50 GiB)


This is isabelle build -o parallel_proofs=0 to shrink the heap a bit more.

Timing did not change significantly:

(1)
Finished HOL (0:06:13 elapsed time, 0:10:31 cpu time, factor 1.69)
Finished HOL-Library (0:04:11 elapsed time, 0:11:14 cpu time, factor 2.69)
Finished HOL-Computational_Algebra (0:01:14 elapsed time, 0:02:22 cpu time, 
factor 1.91)
Finished HOL-Algebra (0:04:39 elapsed time, 0:10:42 cpu time, factor 2.30)
Finished Padic_Ints (0:03:23 elapsed time, 0:07:18 cpu time, factor 2.16)
Finished Padic_Field (0:12:49 elapsed time, 0:21:32 cpu time, factor 1.68)

(2)
Finished Pure (0:00:14 elapsed time, 0:00:14 cpu time, factor 0.97)
Finished HOL (0:06:05 elapsed time, 0:11:26 cpu time, factor 1.88)
Finished HOL-Library (0:03:58 elapsed time, 0:10:30 cpu time, factor 2.65)
Finished HOL-Computational_Algebra (0:01:08 elapsed time, 0:02:11 cpu time, 
factor 1.92)
Finished HOL-Algebra (0:04:20 elapsed time, 0:10:06 cpu time, factor 2.33)
Finished Padic_Ints (0:02:51 elapsed time, 0:05:34 cpu time, factor 1.95)
Finished Padic_Field (0:14:34 elapsed time, 0:24:42 cpu time, factor 1.70)


	Makarius


From lp15 at cam.ac.uk  Wed May 31 12:52:51 2023
From: lp15 at cam.ac.uk (Lawrence Paulson)
Date: Wed, 31 May 2023 10:52:51 +0000
Subject: [isabelle-dev] NEWS: significantly reduced ML heap usage
In-Reply-To: <60de0191-7c63-973c-b907-0168d6e74669@sketis.net>
References: <8ce946ef-f740-2aa3-fddb-87988e5c9a21@sketis.net>
 <60de0191-7c63-973c-b907-0168d6e74669@sketis.net>
Message-ID: <AFC80721-B7EA-46CF-8C76-60852705E791@cam.ac.uk>

These are tremendous improvements! Many thanks.

As you know, people are relying more and more locales to structure elaborate definition hierarchies, so we need this. And I gather that the users of certain other systems have to put up with major performance issues once a proof is more than a couple of dozen lines long.

Larry

> On 30 May 2023, at 21:10, Makarius <makarius at sketis.net> wrote:
> 
> Here are some more measurements to show the quite significant effect on locale-laden applications.
> 
> This is (1) current Isabelle/f3d19c8445ec + AFP/d34609a6a678 versus (2) Isabelle2022 + afp-2022/2458cc9f2178 on the rather heavy session Padic_Ints.