INTERNAL ERROR: Black produced code that is not equivalent to the source.

[mithrandi]

This is happening with three source files, running black --check . over Axiom:

https://github.com/twisted/axiom/blob/35098ebf2f4a799904ac8a65c12969f89c2e1f10/axiom/scripts/axiomatic.py

--- src
+++ dst
@@ -254,11 +254,11 @@
           body=
             Expr(
               value=
                 Constant(
                   value=
-                    b'Turn a byte string from the command line into a unicode string.\n        ',  # bytes
+                    b'Turn a byte string from the command line into a unicode string.',  # bytes
                 )  # /Constant
             )  # /Expr
             Assign(
               targets=
                 Name(

https://github.com/twisted/axiom/blob/35098ebf2f4a799904ac8a65c12969f89c2e1f10/axiom/attributes.py

--- src
+++ dst
@@ -14618,11 +14618,11 @@
       body=
         Expr(
           value=
             Constant(
               value=
-                b'A type constraint was violated.\n    ',  # bytes
+                b'A type constraint was violated.',  # bytes
             )  # /Constant
         )  # /Expr
         FunctionDef(
           args=
             arguments(
@@ -21657,11 +21657,11 @@
       body=
         Expr(
           value=
             Constant(
               value=
-                b"\n    Attribute representing a number with a specified number of decimal\n    places.\n\n    This is stored in SQLite as a binary integer multiplied by M{10**N}\n    where C{N} is the number of decimal places required by Python. \n    Therefore, in-database multiplication, division, or queries which\n    compare to integers or fixedpointdecimals with a different number of\n    decimal places, will not work.  Also, you cannot store, or sum to, fixed\n    point decimals greater than M{(2**63)/(10**N)}.\n\n    While L{ieee754_double} is handy for representing various floating-point\n    numbers, such as scientific measurements, this class (and the associated\n    Python decimal class) is more appropriate for arithmetic on sums of money.\n\n    For more information on Python's U{Decimal\n    class<http://www.python.org/doc/current/lib/module-decimal.html>} and on\n    general U{computerized Decimal math in\n    general<http://www2.hursley.ibm.com/decimal/decarith.html>}.\n\n    This is currently a private helper superclass because we cannot store\n    additional metadata about column types; maybe we should fix that.\n\n    @cvar decimalPlaces: the number of points of decimal precision allowed by\n    the storage and retrieval of this class.  *Points beyond this number\n    will be silently truncated to values passed into the database*, so be\n    sure to select a value appropriate to your application!\n    ",  # bytes
+                b"\n    Attribute representing a number with a specified number of decimal\n    places.\n\n    This is stored in SQLite as a binary integer multiplied by M{10**N}\n    where C{N} is the number of decimal places required by Python.\n    Therefore, in-database multiplication, division, or queries which\n    compare to integers or fixedpointdecimals with a different number of\n    decimal places, will not work.  Also, you cannot store, or sum to, fixed\n    point decimals greater than M{(2**63)/(10**N)}.\n\n    While L{ieee754_double} is handy for representing various floating-point\n    numbers, such as scientific measurements, this class (and the associated\n    Python decimal class) is more appropriate for arithmetic on sums of money.\n\n    For more information on Python's U{Decimal\n    class<http://www.python.org/doc/current/lib/module-decimal.html>} and on\n    general U{computerized Decimal math in\n    general<http://www2.hursley.ibm.com/decimal/decarith.html>}.\n\n    This is currently a private helper superclass because we cannot store\n    additional metadata about column types; maybe we should fix that.\n\n    @cvar decimalPlaces: the number of points of decimal precision allowed by\n    the storage and retrieval of this class.  *Points beyond this number\n    will be silently truncated to values passed into the database*, so be\n    sure to select a value appropriate to your application!\n    ",  # bytes
             )  # /Constant
         )  # /Expr
         FunctionDef(
           args=
             arguments(

https://github.com/twisted/axiom/blob/35098ebf2f4a799904ac8a65c12969f89c2e1f10/axiom/store.py

--- src
+++ dst
@@ -492,11 +492,11 @@
       body=
         Expr(
           value=
             Constant(
               value=
-                b'You must define some attributes on every item.\n    ',  # bytes
+                b'You must define some attributes on every item.',  # bytes
             )  # /Constant
         )  # /Expr
       decorator_list=
       name=
         'NoEmptyItems',  # str
@@ -35928,11 +35928,11 @@
           body=
             Expr(
               value=
                 Constant(
                   value=
-                    b'For use with SELECT (or SELECT-like PRAGMA) statements.\n        ',  # bytes
+                    b'For use with SELECT (or SELECT-like PRAGMA) statements.',  # bytes
                 )  # /Constant
             )  # /Expr
             If(
               body=
                 Assign(

This happens with master (20.8b2.dev6+g7fe19fa) too.

[myselfhimself]

Hi
There is a discrepancy of formatting between black 19 (local computer) and black 20 (freshest from CI)

I am trying to downgrade to a now locked version of black to prevent this in my CI, but I cannot remember which version black would accept to install on Mac OS with pip (side-issue..)

[ambv]

This is #1632 and #1635, but with Black actually catching it as and AST discrepancy. Somehow the other examples don't.

[myselfhimself]

Hi
There is a discrepancy of formatting between black 19 (local computer) and black 20 (freshest from CI)

I am trying to downgrade to a now locked version of black to prevent this in my CI, but I cannot remember which version black would accept to install on Mac OS with pip (side-issue..)

Now unsure of this. Please ignore my comment for now.

[ichard26]

@ambv for the case of #1635, PR #1593 allows the removal of trailing space in docstrings so that's why the AST checks don't catch it.

[JelleZijlstra]

This no longer crashes on current main.