And of course I jumped the gun because I see #8955 not yet merged.

And of course I jumped the gun because I see #8955 not yet merged.

@taldcroft - You're welcome to merge it, or approve so I can do it. Dropping python 3.5 for 4.0 was agreed a while ago, and it's also in the APE.

About merging, presumably it needs to pass all the CI tests first. It does pass locally on my laptop.

As long as the 3.5 test setup is included, CircleCI will never make it to the 3.6 tests. But the Travis failures all seem to be due to the np.broadcast_arrays DeprecationWarnings from numpy 1.18.

Tested with Python 3.6 + 3.8b1 on my Mac as well.

the crash in astropy/utils/console.py was fixed in asottile/pyupgrade#173 fwiw -- just haven't made a release yet

Here's the diff if you want to apply it:

$ git diff | cat
diff --git a/astropy/utils/console.py b/astropy/utils/console.py
index a8040906d..c89da247b 100644
--- a/astropy/utils/console.py
+++ b/astropy/utils/console.py
@@ -256,7 +256,7 @@ def _color_text(text, color):
         return text
 
     color_code = color_mapping.get(color, '0;39')
-    return '\033[{0}m{1}\033[0m'.format(color_code, text)
+    return f'\033[{color_code}m{text}\033[0m'
 
 
 def _decode_preferred_encoding(s):
@@ -425,12 +425,12 @@ def human_time(seconds):
     seconds = int(seconds)
 
     if seconds < 60:
-        return '   {0:2d}s'.format(seconds)
+        return f'   {seconds:2d}s'
     for i in range(len(units) - 1):
         unit1, limit1 = units[i]
         unit2, limit2 = units[i + 1]
         if seconds >= limit1:
-            return '{0:2d}{1}{2:2d}{3}'.format(
+            return '{:2d}{}{:2d}{}'.format(
                 seconds // limit1, unit1,
                 (seconds % limit1) // limit2, unit2)
     return '  ~inf'
@@ -482,10 +482,10 @@ def human_file_size(size):
         str_value = str_value[:2]
     else:
         str_value = str_value[:3]
-    return "{0:>3s}{1}".format(str_value, suffix)
+    return f"{str_value:>3s}{suffix}"
 
 
-class _mapfunc(object):
+class _mapfunc:
     """
     A function wrapper to support ProgressBar.map().
     """
@@ -648,10 +648,10 @@ class ProgressBar:
         else:
             t = ((time.time() - self._start_time) * (1.0 - frac)) / frac
             prefix = ' ETA '
-        write(' {0:>4s}/{1:>4s}'.format(
+        write(' {:>4s}/{:>4s}'.format(
             human_file_size(value),
             self._human_total))
-        write(' ({:>6.2%})'.format(frac))
+        write(f' ({frac:>6.2%})')
         write(prefix)
         if t is not None:
             write(human_time(t))
@@ -685,7 +685,7 @@ class ProgressBar:
         # Calculate percent completion, and update progress bar
         frac = (value/self._total)
         self._widget.value = frac * 100
-        self._widget.description = ' ({:>6.2%})'.format(frac)
+        self._widget.description = f' ({frac:>6.2%})'
 
     def _silent_update(self, value=None):
         pass

I re-did this from scratch using a hacked version of pyupgrade at e477316 to convert format and % strings to modern formatting, including f-strings where possible. The hack bit was to essentially de-select all upgrades except for string formatting ones. That removes (at least from initial quick inspection) all the issues with indentation, and also makes the scope of this PR more contained.

For the record:

diff --git a/pyupgrade.py b/pyupgrade.py
index f048475..960c966 100644
--- a/pyupgrade.py
+++ b/pyupgrade.py
@@ -411,7 +411,7 @@ def _fix_py2_compatible(contents_text):
         return contents_text
     visitor = Py2CompatibleVisitor()
     visitor.visit(ast_obj)
-    if not any((
+    if True or not any((
             visitor.dicts,
             visitor.sets,
             visitor.set_empty_literals,
@@ -1515,7 +1515,7 @@ def _fix_py3_plus(contents_text):
     visitor = FindPy3Plus()
     visitor.visit(ast_obj)
 
-    if not any((
+    if True or not any((
             visitor.bases_to_remove,
             visitor.encode_calls,
             visitor.if_py2_blocks,

@taldcroft - can the tool understand the narrative docs too? if yes, could you run in on the code snippets in there, either as part of this PR or a separate one?

@bsipocz - I'd rather leave the narrative docs as a separate PR. 255 files is already big enough, and some custom code would be needed for the narrative docs as the tool is expecting only parsable Python code.

There is already conflict. I'd say, either we get this in soon, or break it into smaller pieces by submodule.

From experience you'll frustrate contributors less by landing it all in one PR (especially if patches tend to span multiple modules) as they'll only have to deal with conflicts once

I'm happy to hit merge once this is rebased and agree that now that it's done, we don't need to split up to multiple PRs (e.g. I'm happy to help sort out other PRs that get conflicts due to this).

OK, this is now re-processed using current master.

Once more... and in the interest of reproducibility here is what I typed (laying bare my unsophisticated approach):

cd ..
find astropy/astropy -type f -name '*.py' | 
  grep -v astropy/astropy/extern/  | 
  grep -v astropy/astropy/modeling/core.py | 
  xargs python ~/git/pyupgrade/pyupgrade.py --py36-plus

OK, well, I take the blame if this messes something up due to the merging, but based on the 5 rockets, I go ahead and merge now.

@taldcroft - should the recommendation in codeguide.rst be updated accordingly to put

for tea="spam" f"{tea:s}"

in first place?