Why it causes core dump

[wqh17101]

Details

• Read the Docs project URL:https://readthedocs.org/projects/matchzoo/
• Build URL (if applicable):https://readthedocs.org/projects/matchzoo/builds/7406071/ and https://readthedocs.org/projects/matchzoo/builds/77406100/
• Read the Docs username (if applicable):wqh17101

Expected Result

Get the right content which generated by sphinx-apidoc

Actual Result

The 77406100 is the one can build normally. But it can not display the content generated by sphinx-apidoc ,so i check the log

i find an error , obviously, it caused the bug.
So i fixed it and commit the file .
it works in my local env

but in the readthedoc's env ,it got the result of 7406071

i got a core dump
And it seems that whatever i change (the right code in my env )and commit ,it would usually get a core dump.

[wqh17101]

I think there maybe something wrong i did.
The first time i build, it worked. And never worked again.
What i did is change the A file,and commit the A file then build online. The A file is the source of my doc's content
i wonder whether i should commit not only the A file,but also the build and source dir which generated from A file.

[stsewd]

You reported this already in #4301, right? Can we keep the conversation just in one issue? or I misunderstood and those are different problems?

[wqh17101]

They are two problems . #4301 is about py36 support .
And this one is about core dump in py35 @stsewd

[stsewd]

I think this has nothing to do with the version, 3.6 is supported. I'm running your project locally to see what is happening.

[stsewd]

Sorry, I can't replicate this in my local installation. But your latest build is passing https://readthedocs.org/projects/matchzoo/builds/7407005/, were you able to solve the problem?

[wqh17101]

No,i reset the repo @stsewd

[wqh17101]

what do you need to replicate ? @stsewd

[stsewd]

@wqh17101

No,i reset the repo

What do you mean? Did you revert a commit?

what do you need to replicate ?

The core dump error in my rtd local installation.

[wqh17101]

@stsewd
What do you mean? Did you revert a commit?
yes
i just want to ask you what i can offer you to replicate the problem

[wqh17101]

should i change the commit to the situation that the problem happened?

[stsewd]

@wqh17101 You can just link the commit here and I try that in my local instance, no need to revert your project

[bwanglzu]

@stsewd I guess the latest commit is this one: link.

[bwanglzu]

@wqh17101 Check this link, downgrade your tensorflow to 1.5 .

[wqh17101]

This is my teammate @bwanglzu ,you can communication with him. @stsewd

[stsewd]

#4308 (comment) this means it is a tensorflow bug? If so, we can close this.

I can't replicate with this commit NTMC-Community/MatchZoo@2ba5977

[bwanglzu]

@stsewd I have no clue. why it raises when building documentations?..I'm not able to replicate as well. Maybe the best thing to do is to wait until tomorrow (he went to sleep just now).

I'll ask him to downgrade tensorflow, if it works we'll close this issue.

[humitos]

Do you need tensorflow to build the docs? I'd suggest you to use a different requirement.txt file to build your docs with only what's strictly necessary on it.

[bwanglzu]

@humitos Apparently we do not need tensoflow to build the docs (we didn't even have requirements.txt for documentation) :(

[humitos]

RTD auto pick the requirements in the root of your project. That's why.

You may want to check http://docs.readthedocs.io/en/latest/yaml-config.html and http://docs.readthedocs.io/en/latest/yaml-config.html and specify a reqs file there.

[stsewd]

we didn't even have requirements.txt for documentation

oh, I see. RTD tries to find a requirements.txt automatically in the root of the repo, you can create an empty docs/requirements.txt.

We are tracking the other bug here #1621

[bwanglzu]

@humitos @stsewd That make sense, thanks! I'll create a requirements.txt under docs folder and try to re-build it again, thanks!

[stsewd]

@bwanglzu let us know how that goes!

[bwanglzu]

@stsewd Thanks! Will reply tomorrow!

[wqh17101]

It doesn't work. @stsewd The core problem is that why i can build at the first time and when i make some right commits ,it failed. There is no requirements change between the origin and the commit

you can see this one https://readthedocs.org/projects/matchzoo/builds/7408927/ i downgrade the tensorflow.
Apparently if there are some bugs in the requirements,i will never succeed. But i succeed to build from the origin one

local build always succeed. :)

[stsewd]

I built your project several times in my local instance and I wasn't able to replicate this. So I tested it in the readthedocs.org site, I was able to replicate there... but is weird, I'm using the 2.0 docker image in my local instance @humitos which one is used by rtd in production?

@wqh17101 can you test creating a readthedocs.yml file with this:

build:
  image: 2.0

python:
  version: 3.5

[wqh17101]

Let's compare the result.
It seems that it crashed when reading sources matchzoo.engine,where the file i commit is.
@stsewd

[wqh17101]

@wqh17101 can you test creating a readthedocs.yml file with this:
i will try @stsewd

[wqh17101]

it doesn't work @stsewd . So i think it is not the problem about setting of readthedocs. We should focus on why i modify the file in matchzoo.engine,it will cause reading source failure on the site @stsewd

[stsewd]

Can you please point me to that file in your repo?

[wqh17101]

Sure , the file is https://github.com/faneshion/MatchZoo/blob/feature/document/matchzoo/engine/param_table.py

And the origin one is that i comment which can build normally.But the build env is 3.5 ,so it doesn't support f-string, so i just change the usage to present the same msg and get a core dump.
@stsewd

[wqh17101]

Something new? @stsewd

[stsewd]

Sorry, no. I can't replicate locally so I don't know how to help.

[wqh17101]

OK i will try whether i remain the F-string,i can do everything
Remain the F-string , modify the docstring, i can build
Remain the F-string , modify the other codes, i can build
Remain the F-string , modify the content in F-STRING , i can build
Remain one F-string , i can build
remain no F-string,core dump
remain F-STRING,using 3.6 to build by yml ,core dump https://readthedocs.org/projects/matchzoo/builds/7416075/
yml:

build:
  image: latest
python:
  version: 3.6

remain F-STRING,using 3.5 to build by yml ,i can build https://readthedocs.org/projects/matchzoo/builds/7416094/
yml:

build:
  image: latest

python:
  version: 3.5

what does the f do? It is amazing. Remove it ,it causes core dump, whatever i commit.
@stsewd

[wqh17101]

Do my make and makefile have problem? @stsewd https://github.com/faneshion/MatchZoo/blob/feature/document/matchzoo/docs

[wqh17101]

Can I login the auto build env

[stsewd]

Do my make and makefile have problem?

Readthedocs uses sphinx-build command to build your documentation, so it has nothing to do with your makefile.

[wqh17101]

@stsewd with tensorflow==1.4.0 in my requirements, it works well. I think you should communicate with Tensorflow with this bug

[stsewd]

@wqh17101 thanks for the information, this is related to #3738 (see #3738 (comment)). Also, this is a tensorflow bug, closing then.