Can anyone see if they can reproduce a bug on their computer? It wont take long.

New comment
 

I have been experiencing a lot of bugs in the forward tester. One is that if there are more than 65536 passes to do in the forward tester, right at 65536 it stops finishing passes. I get an error that says xx forward passes not processed and returned to task queue. If you try to stop it or close it the program freezes and I haven't ever seen it unfreeze. I have tested this on 2 different computers with the same result. I am wondering if anyone else would experience this issue. If anyone can try running this ea on your computer I would appreciate it. It would only take a couple minutes. The settings are EURUSD, 1M, the time period is custom set to the last 7 days, with forward testing at 1/4, Zero delay, 1 minute OLHC, fast genetic based algorithm, and custom max. Every input is set to start 0, step 1, and end 100,000.

input int test1;
input int test2;
input int test3;
input int test4;
input int test5;
input int test6;
input int test7;
input int test8;
input int test9;
input int test10;
input int test11;
input int test12;
input int test13;
input int test14;
input int test15;
input int test16;
input int test17;
input int test18;
input int test19;
input int test20;
int OnInit()
  {
   return(INIT_SUCCEEDED);
  }
double OnTester()
  {
   return(test1+test2+test3+test4+test5+test6+test7+test8+test9+test10+test11+test12+test13+test14+test15+test16+test17+test18+test19+test20);
  }
 
tanner gilliland:

I have been experiencing a lot of bugs in the forward tester. One is that if there are more than 65536 passes to do in the forward tester, right at 65536 it stops finishing passes. I get an error that says xx forward passes not processed and returned to task queue. If you try to stop it or close it the program freezes and I haven't ever seen it unfreeze. I have tested this on 2 different computers with the same result. I am wondering if anyone else would experience this issue. If anyone can try running this ea on your computer I would appreciate it. It would only take a couple minutes. The settings are EURUSD, 1M, the time period is custom set to the last 7 days, with forward testing at 1/4, Zero delay, 1 minute OLHC, fast genetic based algorithm, and custom max. Every input is set to start 0, step 1, and end 100,000.

Ok, I will run the test as you ask and report back when done!
 

Here are my results for the test ...

The setup ...


The optimisation phase ...

During the forward test phase, it got stuck at 65536, and stopped processing (finished), as described by the OP.


However, after a few minutes the agents started up again and the Journal continued to show activity ...

After a while it stopped again, but so far nothing has frozen and MetaTrader is still responsive. This may be due to me only using 4 out of the 8 threads for the agents, allowing for the other threads to continue to work normally.

Every now and then the CPU spikes and MetaTester shows activity.

I will leave it running and see how it progresses and report back later.

 

The journal is still showing activity even though the count is stuck at 65536 of 95507. But it is only doing it in spurts, not continuously ...


 

However, it seems to be repeating things in an endless cycle ...

I can see that many of the forward tests are being repeated ...


 

So, at this point I clicked on [Stop] and it froze, just as reported by the OP ...

 

I have attached the complete copy of the Strategy Tester logs for review ... and the Settings (and inputs) file ...

 

Thank you for taking the time to replicate the bug. If you have time and wouldn't mind I was wondering if you can try to replicate 2 more bugs. They are less critical than the 65536 bug but I would still appreciate it if you took a look. You should be able to see them using the same setup as before. The only thing I would change is the timeframe, to 4-8 months. Its very quick and hard to see with only 7 days. I see the bug happen great in my optimizations that take up to a week long, but with the test setup at 4-8 months you can still see it and it only takes about 10 minutes. If you want to see the bugs even better you could increase the timeframe even more. I will attach a screen recording in case my post is hard to understand. (I can't explain things well) The first bug happens at either 256, 512, or 768 passes, depending on how many total passes there are. I assume it can happen past that at every multiple of 256, but I haven't ever seen it higher than 768. 512 is the most common one, so I will just use that as the number from now on. For the first bug, during the first 512 passes there are more passes being completed by the agents/cores than what is shown in the number completed at the bottom. When the forward testing starts it looks like they line up. However, as the passes completed number gets closer to 512 the total completed number will start to slow down, even though the rate of the passes completed by the agents/cores is the same. To see this, click on the agents tab right when forward testing starts. Keep watch of the passes completed by agents and the total passes completed number at the bottom. The total passes will start off fast then slow down when it gets closer to 512. The agents will complete cores at a constant pace but when they complete a pass it will not get added to the total passes number. The second bug happens right after the first and happens until double the 256,512, or 768 number of the first bug. So 256-512, 512-1024, or 768-1536. The second bug does the same thing as the first. It slows down when it gets closer to 1024 even though the passes are being completed at the same rate. However, it works on passes that should be at the end of the graph. I can't describe this well, but I will attach a screenshot of a forward test where it is more apparent. It will look like you could take the passes from 512 to 1024, put them at the end of the graph and only then will the graph look correct. I had previously posted about these 2 bugs, the bug where it gets frozen at 65536, and one more where the forward results tab doesn't match single test results in this badly written post (https://www.mql5.com/en/forum/454524). I would have wanted to see if you could replicate the last bug, but Alain Verleyen already tried and was not able to. So I need to see if something is wrong with the setup or if I am doing something wrong still. If you don't have time to test any of this no worries. I am glad you were able to do reproduce one of the more annoying bugs. I also believe they are related and hopefully when they go to fix the 65536 bug the other bugs might get fixed as well. Thanks.

Link to screen capture in my google drive: https://drive.google.com/file/d/1DFV27IpoMqrYSvBd1s4I0_s1GK-L-j7q/view?usp=sharing It is still super fast and hard to see but you can still see it stop at 512 and 1024 and you can see how the passes from 512 to 1024 are passes that should be at the end.

MT5 forward testing bugs. - Here's how forward testing bugs affect your results
MT5 forward testing bugs. - Here's how forward testing bugs affect your results
  • 2023.09.23
  • www.mql5.com
This post is about a some weirdly related (i think) bugs that happen when forward testing optimized results. The first bug is that the displayed results from the top section of backtesting passes in forward tab are wrong
 
tanner gilliland #: Thank you for taking the time to replicate the bug. If you have time and wouldn't mind I was wondering if you can try to replicate 2 more bugs. They are less critical than the 65536 bug but I would still appreciate it if you took a look. You should be able to see them using the same setup as before. The only thing I would change is the timeframe, to 4-8 months. Its very quick and hard to see with only 7 days. I see the bug happen great in my optimizations that take up to a week long, but with the test setup at 4-8 months you can still see it and it only takes about 10 minutes. If you want to see the bugs even better you could increase the timeframe even more. I will attach a screen recording in case my post is hard to understand. (I can't explain things well) The first bug happens at either 256, 512, or 768 passes, depending on how many total passes there are. I assume it can happen past that at every multiple of 256, but I haven't ever seen it higher than 768. 512 is the most common one, so I will just use that as the number from now on. For the first bug, during the first 512 passes there are more passes being completed by the agents/cores than what is shown in the number completed at the bottom. When the forward testing starts it looks like they line up. However, as the passes completed number gets closer to 512 the total completed number will start to slow down, even though the rate of the passes completed by the agents/cores is the same. To see this, click on the agents tab right when forward testing starts. Keep watch of the passes completed by agents and the total passes completed number at the bottom. The total passes will start off fast then slow down when it gets closer to 512. The agents will complete cores at a constant pace but when they complete a pass it will not get added to the total passes number. The second bug happens right after the first and happens until double the 256,512, or 768 number of the first bug. So 256-512, 512-1024, or 768-1536. The second bug does the same thing as the first. It slows down when it gets closer to 1024 even though the passes are being completed at the same rate. However, it works on passes that should be at the end of the graph. I can't describe this well, but I will attach a screenshot of a forward test where it is more apparent. It will look like you could take the passes from 512 to 1024, put them at the end of the graph and only then will the graph look correct. I had previously posted about these 2 bugs, the bug where it gets frozen at 65536, and one more where the forward results tab doesn't match single test results in this badly written post (https://www.mql5.com/en/forum/454524). I would have wanted to see if you could replicate the last bug, but Alain Verleyen already tried and was not able to. So I need to see if something is wrong with the setup or if I am doing something wrong still. If you don't have time to test any of this no worries. I am glad you were able to do reproduce one of the more annoying bugs. I also believe they are related and hopefully when they go to fix the 65536 bug the other bugs might get fixed as well. Thanks. Link to screen capture in my google drive: https://drive.google.com/file/d/1DFV27IpoMqrYSvBd1s4I0_s1GK-L-j7q/view?usp=sharing It is still super fast and hard to see but you can still see it stop at 512 and 1024 and you can see how the passes from 512 to 1024 are passes that should be at the end.

Regarding your original issues, you can see how I replicated and reported every step. I include screenshots, brief explanations and copies of the journal and other logs.

What I presented, you could have easily done yourself to report it to the admins or developers. So, now regarding this recent issue, I leave it up to you to do something similar.

Also, please don’t write such a large contiguous block of text, with no breaks and without proper paragraphing. It makes it impossible to read and digest. I gave up reading it after the first couple of lines.

Break it down into a logical, organised outline. Then fill in each section with the details, but keep it concise.

Once you have properly organised your logic and presented your issue, then post a brief introduction in the relevant MetaQuotes topic and link it to your full report here.

 
tanner gilliland:

I have been experiencing a lot of bugs in the forward tester. One is that if there are more than 65536 passes to do in the forward tester, right at 65536 it stops finishing passes. I get an error that says xx forward passes not processed and returned to task queue. If you try to stop it or close it the program freezes and I haven't ever seen it unfreeze. I have tested this on 2 different computers with the same result. I am wondering if anyone else would experience this issue. If anyone can try running this ea on your computer I would appreciate it. It would only take a couple minutes. The settings are EURUSD, 1M, the time period is custom set to the last 7 days, with forward testing at 1/4, Zero delay, 1 minute OLHC, fast genetic based algorithm, and custom max. Every input is set to start 0, step 1, and end 100,000.

You are lucky that Fernando tried your test.

You need to understand that developers are very busy and your way to write long text instead of providing files and screenshot is a no go.

1. There is no EA file attached, there is a need to copy and paste, create a new file, etc... MQ developers will not even do it. Your "report" is already done.

2. The Strategy Tester settings in "words". Second kill, you can provide an ini file with these settings are at least a screenshot. Second kill.

3. The EA settings, 20 inputs to set manually, report killed a third time. You need to provide a set file with the settings.

MQ developers will not even check, I can guarantee that to you.

As Fernando took the time to reproduce it, and as I have now some spare time I will check nonetheless. But I am not happy I have to do these 3 things manually.

 
@Alain Verleyen #:

2. The Strategy Tester settings in "words". Second kill, you can provide an ini file with these settings are at least a screenshot. Second kill.

3. The EA settings, 20 inputs to set manually, report killed a third time. You need to provide a set file with the settings.

... and as I have now some spare time I will check nonetheless. But I am not happy I have to do these 3 things manually.

To help you out, I added my Settings (.ini) file to my post
12
New comment
Reason: