Single term repeat using up all Workspace

[nikolai-husung]

The following (admittedly dumb) example runs quickly out of Workspace (around -D POWER=1000). The commented variant runs just fine.

Symbol x,n;
Local test = 5*x^`POWER';
repeat id x^n?pos_ = n*x^(n-1);
*#do dummy=1,`POWER'
*   id x^n?pos_ = n*x^(n-1);
*#enddo
.sort
Print +s;
.end

Since there already is a check whether one runs out of Workspace, a quick fix might be to first check if the Workspace is saturated with no-longer needed terms? This should not slow down code that would otherwise run through just fine, while reducing the pressure on Workspace somewhat. For repeat generating new terms this limit will be reached even quicker.

[vermaseren]

As your example is given, there is a fundamental problem with your proposal. You might not expand the workspace, if you know for sure that the rhs will never generate more than one term. But in your example the (n-1) is a subexpression, the expansion has to go through TestSub and through Generator again and to see that this eventually collapses would require quite some code, and hence slow things down. Things would already be more favourable if you would have written x^n/x because that contains no subexpressions. You will still have to have something in Generator that tells Form that there is only one term, but that code would be significantly simpler. The best way would be, no matter what, to use combinatorics functions to avoid runaway recursions, or limit such recursions to a limited depth. Personally I am against sacrificing even small amounts of speed to run some benchmarks faster, when intervention of the user can avoid running into such cases. The combinatorics functions can catch a lot of this stuff, and often make your program much faster anyway.

…

On 10 Jul 2026, at 17:30, Nikolai Husung ***@***.***> wrote: nikolai-husung created an issue (form-dev/form#873) <#873> The following (admittedly dumb) example runs quickly out of Workspace (around -D POWER=1000). The commented variant runs just fine. Symbol x,n; Local test = 5*x^`POWER'; repeat id x^n?pos_ = x^(n-1); *#do dummy=1,`POWER' * id x^n?pos_ = n*x^(n-1); *#enddo .sort Print +s; .end Since there already is a check whether one runs out of Workspace, a quick fix might be to first check if the Workspace is saturated with no-longer needed terms? This should not slow down code that would otherwise run through just fine, while reducing the pressure on Workspace somewhat. For repeat generating new terms this limit will be reached even quicker. — Reply to this email directly, view it on GitHub <#873?email_source=notifications&email_token=ABJPCERMJBREDG7H675A72D5EED3DA5CNFSL4Z3JMQ5C6L3HNF2C22DVMIXUS43TOVSS6NBYGU3DGOBRGMZDJJTSMVQXG33OVJZXKYTTMNZGSYTFMSSWK5TFNZ2KYZTPN52GK4S7MNWGSY3L>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABJPCEUBBCTO3MSR3GT3KGL5EED3DAVCNFSNUABEKJSXA33TNF2G64TZHMYTENZVGYZDGMB3JFZXG5LFHM2DQNJWGM4DCMZSGSQXMAQ>. Triage notifications, keep track of coding agent tasks and review pull requests on the go with GitHub Mobile for iOS <https://github.com/notifications/mobile/ios/ABJPCEWKMBU3XGQ3FM4USHT5EED3DA5CNFSL4Z3JMQ5C6L3HNF2C22DVMIXUS43TOVSS6NBYGU3DGOBRGMZDJJTSMVQXG33OVJZXKYTTMNZGSYTFMSSWK5TFNZ2KUZTPN52GK4S7NFXXG> and Android <https://github.com/notifications/mobile/android/ABJPCEWNBVLPP62YRRSSKAD5EED3DA5CNFSL4Z3JMQ5C6L3HNF2C22DVMIXUS43TOVSS6NBYGU3DGOBRGMZDJJTSMVQXG33OVJZXKYTTMNZGSYTFMSSWK5TFNZ2K4ZTPN52GK4S7MFXGI4TPNFSA>. Download it today! You are receiving this because you are subscribed to this thread.

[tueda]

The following change was known to reduce workspace usage (at least 18 years ago), but no one knew what side effects it might have:

diff --git a/sources/proces.c b/sources/proces.c
index 06df6280..82186f42 100644
--- a/sources/proces.c
+++ b/sources/proces.c
@@ -3992,7 +3992,7 @@ CommonEnd:
 */
                        }
                } while ( ( i = TestMatch(BHEAD term,&level) ) == 0 );
-               if ( AT.WorkPointer < term + *term ) AT.WorkPointer = term + *term;
+               AT.WorkPointer = term + *term;
                if ( i > 0 ) replac = TestSub(BHEAD term,level);
                else replac = i;
                if ( replac >= 0 || AT.TMout[1] != SYMMETRIZE ) {

So, I asked GPT-5.6 about it. Here is the report. If we choose to believe it, we can fix the workspace usage problem.

[jodavies]

Targeted runs also exercised recursive identities, identities inside function arguments, transforms, argument processing, contraction, tracing and symmetrisation before the complete suites were run.

What does this mean? Did it generate and run tests?

[tueda]

Targeted runs also exercised recursive identities, identities inside function arguments, transforms, argument processing, contraction, tracing and symmetrisation before the complete suites were run.

What does this mean? Did it generate and run tests?

I don't think it generated any new tests, but I think it used existing tests.

GPT-5.6 tried to run several tests (including at least the reproducer in section 4, and test cases NotesMetric_1, Transform-mulargs_1 and Sta_Argument_2) in debugger-assisted runs that simulated the unconditional substitution of AT.WorkPointer at line 3995 (I initially didn't allow any changes to the source code). The sentence "Targeted runs also exercised ..." seems to refer to this.

Later, I suggested modifying the code and running the test cases. Then it ran:

check/check.rb build/sources/form
check/check.rb -w 4 build/sources/tform

[vermaseren]

There was, in the past, a reason why it was set up the way it is currently. It may be, that my memory fails here, but if you look in TestSub, it can in principle make terms longer. This was a big p[ain in the neck, and it took a long time before all bugs were removed that were related to this. If you want to follow the suggestions of the AI, you should first design some tests that make use of the cases in TestSub. It may well be that none of the current tests do anything critical there, but if you change the code, and you will run in such a bug, you will have a very nasty time debugging. But because this is all many years ago, it may well be, that some other things have been changed, and there is no problem. I cannot tell.

…

On 11 Jul 2026, at 09:43, Takahiro Ueda ***@***.***> wrote: tueda left a comment (form-dev/form#873) <#873 (comment)> Targeted runs also exercised recursive identities, identities inside function arguments, transforms, argument processing, contraction, tracing and symmetrisation before the complete suites were run. What does this mean? Did it generate and run tests? I don't think it generated any new tests, but I think it used existing tests. GPT-5.6 tried to run several tests (including at least the reproducer in section 4, and test cases NotesMetric_1, Transform-mulargs_1 and Sta_Argument_2) in debugger-assisted runs that simulated the unconditional substitution of AT.WorkPointer at line 3995 (I initially didn't allow any changes to the source code). The sentence "Targeted runs also exercised ..." seems to refer to this. Later, I suggested modifying the code and running the test cases. Then it ran: check/check.rb build/sources/form check/check.rb -w 4 build/sources/tform — Reply to this email directly, view it on GitHub <#873?email_source=notifications&email_token=ABJPCEULVJFPECMNHCUT6735EHV2BA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTIOJUGM3DKNBZHA2KM4TFMFZW63VHMNXW23LFNZ2KKZLWMVXHJLDGN5XXIZLSL5RWY2LDNM#issuecomment-4943654984>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABJPCEWHAF2YCFTPLIWB5MD5EHV2BAVCNFSNUABEKJSXA33TNF2G64TZHMYTENZVGYZDGMB3JFZXG5LFHM2DQNJWGM4DCMZSGSQXMAQ>. Triage notifications, keep track of coding agent tasks and review pull requests on the go with GitHub Mobile for iOS <https://github.com/notifications/mobile/ios/ABJPCETITDCKRSFWVTK55GL5EHV2BA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTIOJUGM3DKNBZHA2KM4TFMFZW63VHMNXW23LFNZ2KKZLWMVXHJKTGN5XXIZLSL5UW64Y> and Android <https://github.com/notifications/mobile/android/ABJPCESCIMGY3NVYS525VGT5EHV2BA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTIOJUGM3DKNBZHA2KM4TFMFZW63VHMNXW23LFNZ2KKZLWMVXHJLTGN5XXIZLSL5QW4ZDSN5UWI>. Download it today! You are receiving this because you commented.