dict(row) causing TypeError: tuple indices must be integers or slices, not str

[lindycoder]

DISCLAIMER: This is a very weird bug, I don't know if SQLAlchemy is the problem, but I couldn't reproduce without it, there is a very easy workaround, but I felt like i should report this as it may hide something else.

Describe the bug
Using dict() on an sqlalchemy.engine.Row object is producing TypeError: tuple indices must be integers or slices, not str in a very specific setup (see To Reproduce)

Expected behavior
dict(row) should return the row as a dict.

To Reproduce
This Dockerfile can show the error:

FROM python:3-alpine

RUN apk add --no-cache g++

RUN pip install --upgrade pip && pip install greenlet

RUN apk del g++

RUN pip install sqlalchemy

RUN echo $'\
from sqlalchemy import create_engine \n\
from sqlalchemy.orm import sessionmaker \n\
\n\
session = sessionmaker(bind=create_engine("sqlite://"))() \n\
result = session.execute("select 1, 2, 3").fetchall() \n\
row = result[0] \n\
\n\
print(dict(row)) \n\
' > script.py

RUN python script.py

Inline code is

from sqlalchemy import create_engine 
from sqlalchemy.orm import sessionmaker 

session = sessionmaker(bind=create_engine("sqlite://"))() 
result = session.execute("select 1, 2, 3").fetchall() 
row = result[0] 

print(dict(row))

Error

Traceback (most recent call last):
  File "//script.py", line 8, in <module>
    print(dict(row))
  File "/usr/local/lib/python3.9/site-packages/sqlalchemy/engine/row.py", line 104, in __getitem__
    return self._data[key]
TypeError: tuple indices must be integers or slices, not str

Versions.

• OS: Alpine 3.13
• Python: 3.9.4
• SQLAlchemy: 1.4.6
• Database: sqlite
• DBAPI: ?

Additional context

The provided Dockerfile is a shrunk down version of what we use:

• A base image with build dependencies installing all libs requiring compilation (greenlet)
• A service image that install extra libs not requiring dependencies (sqlalchemy)

I have noticed that REMOVING or MOVING the RUN apk del g++ AFTER the installation of SQL Alchemy will solve the problem.

I also tried to reproduce in debian, compiling libs instead of wheels. with this Dockerfile and could not reproduce, so maybe it's related to Alpine OR there's something else installed in slim-buster.

FROM python:3-slim-buster

RUN apt-get update && apt-get install -y g++

RUN pip install --upgrade pip && pip install greenlet --no-binary :all:

RUN apt-get remove -y g++

RUN pip install sqlalchemy --no-binary :all:

RUN echo '\
from sqlalchemy import create_engine \n\
from sqlalchemy.orm import sessionmaker \n\
\n\
session = sessionmaker(bind=create_engine("sqlite://"))() \n\
result = session.execute("select 1, 2, 3").fetchall() \n\
row = result[0] \n\
\n\
print(dict(row)) \n\
' > script.py

RUN python script.py

WORKAROUND
Using row._asdict() will solve the problem.

Have a nice day! and thank you for your amazing work.

[jvanasco]

I haven't been able to recreate this either, this looks related to your environment. Maybe there is something in the compile args for Python in the image? Alpine images can often be odd, and are famous for being hard to debug.

[lindycoder]

You cannot recreate with the alpine dockerfile?

[jvanasco]

@lindycoder Sorry, like you I could not recreate this on any other environment. I tried a few ubuntu and osx variants. I left the triage label on this, so other contributors will take a look.

[lindycoder]

@jvanasco okay i understand

The fact that NOT removing g++ fixes the problem might be an indication of an installation thing. Is SQL Alchemy doing things differently if a compiler is available?

[CaselIT]

It may be a bug in the python version of the library.

Can you reproduce by installing it on debian (or other) with DISABLE_SQLALCHEMY_CEXT=1 pip install --no-binary sqlalchemy?
This will install slqlachemy without the compiled extension.

The reason that on debian and other system install with compiled extension is because we publish wheel with them active

[zzzeek]

we should also keep in mind that dict(row) is deprecated in any case. you can work around this now by using row._mapping.

[jvanasco]

Is SQL Alchemy doing things differently if a compiler is available?

@lindycoder There are some c-extensions.

Can you reproduce by installing it on debian (or other) with DISABLE_SQLALCHEMY_CEXT=1 pip install --no-binary sqlalchemy?

@CaselIT What about testing on alpine with g++ and disabling cext?

[lindycoder]

we should also keep in mind that dict(row) is deprecated in any case. you can work around this now by using row._mapping.

@zzzeek We started using this a couple weeks ago when we upgraded to a recent version of SQLAlchemy and this stopped working dict(row.items()).

I could work around using row._asdict() which returns _mapping :)

If this is deprecated, what is the official way of translating a row to a dict?

[CaselIT]

@CaselIT What about testing on alpine with g++ and disabling cext?

alpine will download the source distribution, then if it finds the compile it will compile the c-extensions, if it does not find the compile no c-extension is compile. the evn variable DISABLE_SQLALCHEMY_CEXT=1 disables compiling the c-extension even if the compiler is present

[CaselIT]

I could work around using row._asdict() which returns _mapping :)

_asdict is the official way. The idea is that in sa 1.4+ row behaves like a named tuple with the same underscore methods https://docs.python.org/3/library/collections.html#namedtuple-factory-function-for-tuples-with-named-fields

[lindycoder]

@jvanasco @CaselIT
I CAN reproduce on debian AND alpine with g++ installed with DISABLE_SQLALCHEMY_CEXT=1

Alpine

FROM python:3-alpine

RUN apk add --no-cache g++

RUN pip install --upgrade pip && pip install greenlet

RUN DISABLE_SQLALCHEMY_CEXT=1 pip install sqlalchemy

RUN echo $'\
from sqlalchemy import create_engine \n\
from sqlalchemy.orm import sessionmaker \n\
\n\
session = sessionmaker(bind=create_engine("sqlite://"))() \n\
result = session.execute("select 1, 2, 3").fetchall() \n\
row = result[0] \n\
\n\
print(dict(row)) \n\
' > script.py

RUN python script.py

Debian

FROM python:3-slim-buster

RUN apt-get update && apt-get install -y g++

RUN pip install --upgrade pip && pip install greenlet --no-binary :all:

RUN DISABLE_SQLALCHEMY_CEXT=1 pip install sqlalchemy --no-binary :all:

RUN echo '\
from sqlalchemy import create_engine \n\
from sqlalchemy.orm import sessionmaker \n\
\n\
session = sessionmaker(bind=create_engine("sqlite://"))() \n\
result = session.execute("select 1, 2, 3").fetchall() \n\
row = result[0] \n\
\n\
print(dict(row)) \n\
' > script.py

RUN python script.py

[zzzeek]

we should also keep in mind that dict(row) is deprecated in any case. you can work around this now by using row._mapping.

@zzzeek We started using this a couple weeks ago when we upgraded to a recent version of SQLAlchemy and this stopped working dict(row.items()).

I could work around using row._asdict() which returns _mapping :)

If this is deprecated, what is the official way of translating a row to a dict?

_asdict() or _mapping

https://docs.sqlalchemy.org/en/14/core/connections.html#sqlalchemy.engine.Row._mapping