Differences Between Theorems Modules (AMS and non AMS)

[Jean-Marc Lasgouttes]

Le 14/09/2022 à 02:28, Udicoudco a écrit :
> Hello Lyx Developers,
> 
> Recently I've been trying to understand more thoroughly how theorems 
> modules are implemented in LyX, and I think I've come across some bugs, 
> and some redundant modules (but I might have misunderstood a few things).

Hello Udi,

Thanks for looking at that.

> The first thing I've noticed is that all AMS theorems related modules 
> load the packages amsmath and amsthm even if the use did not use any 
> layout.  I could not understand why, as amsthm does not depend on 
> amsmath.  I've tried to compile a lyx file containing all the layouts 
> defined in Theorem (AMS), but i erased the line "Require amsmath,amsthm" 
> from the module, and the file compiled (as amsthm is loaded by the 
> layouts, there is no really a need to load the package if the user did 
> not use any layout).

This looks like a bug indeed.

> The second thing I've noticed, there is no difference between the files 
> theorems.inc and theorems-ams.inc. In the description of the file 
> theorems-std.module, which uses theorems.inc it is said that the module 
> "Defines some theorem environments for use with non-AMS classes", but 
> the module uses amsthm, and he is compilable with AMS classes.

This is indeed surprising. In the old days, non-ams theorem modules did 
not use amsthm. But for some reason it was changed when modularizing the 
layouts. I am not sure whether we can revert this decision 14 years 
later (see below). If we don't, the two types of theorems should be 
merged indeed.

The culprit is:

commit f5a5be4d188a20ee2a5c0213ee11bf64c0908841
Author: Richard Heck <rgheck at comcast.net>
Date:   Thu Jan 10 04:58:20 2008 +0000

     Modularization of the AMS classes, and the theorem environments 
generally.

     This is Part I: The changes to the layout files and modules. Part 
II, still
     to come, will involve lyx2lyx to help people who were using the 
older classes.

> 
> The only difference I've found between theorems.inc and 
> theorems-ams.inc, is the list of layouts the files provide at the 
> beginning of the files (but they provide the same layouts, with the 
> exact same definitions. I've used text comparison software to verify that).
> 
> Because of that, I've concluded that the only differences between 
> theorems-std.module and theorems-ams.module is that in 
> theorems-std.module there are no unnumbered theorems (as 
> theorems-ams.module  includes theorems-starred.inc, but 
> theorems-std.module  does not), theorems-std.module does not load 
> amsmath, and loads amsthm only if the user has used a layout that is not 
> the proof layout, and the definition of the proof layout.

And there is a NextNoIndent difference between the two that if probably 
wrong in one of the instances.

> I did not understand the purpose of the file theorems-proof-std.inc. 
> This file is included in theorems-std.module, and it appears that this 
> file is defining the proof environment (but the definition is quite 
> different from the usual one), if it is not defined. Since amsthm 
> defines the proof environment,  if i will compile a file using Theorems 
> module, the appearance of my proofs will depend if i used another  
> layout provided by the module. Wont it be more reasonable to add to the 
> proof layout defined in theorems-proof.inc the line "Require amsthm" and 
> use this layout? if there is a theorems module that does not use amsthm 
> (and currently i don't think there is one like that), it could include  
> theorems-proof-std.inc.

It made sense when plain theorems did not use amsthm.

> Another problem with the fact that the proof layout in 
> theorems-proof.inc does not require amsthm, is with the module Theorems 
> (Unnumbered). This module only loads amsthm if the user use one of the 
> theorems layouts. if one would simply use the proof layout with this 
> module, and no other layout from the module, with non-AMS class (as 
> these classes also defines the proof environment), he will get the error 
> "LaTeX Error: Environment proof undefined.".

Could you give an example file for that issue?
> 
> Another inquiry I have about the module Theorems (Unnumbered) is the 
> file theorems-starred-equivalents.inc. This file only copies the starred 
> theorems layouts but gives the new layout a name without a star, e.g., 
> it defines a Theorem layout exactly as Theorem* layout (and this files 
> misses the Fact layout).

The comment at the top implies that this sorts the layouts. I do not 
know whether it is still true.

> For all these reasons i believe that with a little bi of modification to 
> the non AMS modules (which actually use amsthm), the AMS modules are 
> redundent, and it is probably wise to add the line "Require amsthm" to 
> the proof layout.
> 
> If this suggestion seems reasonable to everybody, i could start work on 
> the matter, and send the relevent files after editing them as mentioned.

I wonder if we could get rid of theorems.module, and add a theorem-plain 
that use plain LaTeX commands as "theorems.module" used to do. Then 
files using "theorems.module" could be updated by lyx2lyx to use 
theorems-ams instead, and we would have the old functionality back.

> Just for clerification, all the modules files i've mentioned are the 
> ones that comes with LyX2.3.6.1.

It would be nice to check the situation in master, as we are not going 
to fix this (except aybe proof) in 2.3.

Thanks,
JMarc