print, format, and % evolved.

It’s been forty five years since C introduced printf() and the basic formatted printing of positional parameters. Isn’t it time for an upgrade? Yes! Indeed it is.

say goes beyond Python’s print statement/function, format function/method, and % string interpolation operator with simpler, higher-level facilities.

  • DRY, Pythonic templates that piggyback the Python’s well-proven format() method, syntax, and underlying engine.
  • A single output mechanism that works virtually the same in either Python 2 or Python 3 (i.e. seamless compatibility).
  • A companion fmt() object for string formatting.
  • Higher-order line formatting such as line numbering, indentation, and line-wrapping built in. You can get better output formatting with almost no additional code of your own.
  • Convenient methods for common formatting items such as titles, horizontal separators, and vertical whitespace.
  • Easy styled output, including ANSI colors and user-defined styles and text transforms.
  • Easy output to one or more files, again with no additional code.
  • Super-duper template/text aggregator objects for easily building, reading, and writing multi-line texts.


from say import say, fmt

x = 12
nums = list(range(4))
name = 'Fred'

say("There are {x} things.")
say("Nums has {len(nums)} items: {nums}")
say("Name: {name!r}")


There are 12 things.
Nums has 4 items: [0, 1, 2, 3]
Name: 'Fred'

At this level, say is basically a simpler, nicer recasting of:

from __future__ import print_function

print("There are {0} things.".format(x))
print("Nums has {0} items: {1}".format(len(nums), nums))

(The import and numerical sequencing of {} format specs is required to make pure-Python code work correctly from Python 2.6 forward from a single code base.)

Full expressions are are supported within the format braces ({}). Whatever variable names or expressions are found therein will be evaluated in the context of the caller.

The more items that are being printed, and the complicated the format invocation, the more valuable this simple inline specification becomes.

For example:

say("Name: {name:style=blue}", indent='+1')
say("Age:  {age:style=blue}", indent='+1')

Prints a nicely formatted text block, with a proper title and indentation, and just the variable information in blue.

Indentation and Wrapping

Indentation is a common way to display data hierarchically. say will help you manage it. For example:

for item in items:
    say(item, indent=1)

will indent the items by one indentation level (by default, each indent level is four spaces, but you can change that with the indent_str option).

If you want to change the default indentation level:

say.set(indent=1)      # to an absolute level
say.set(indent='+1')   # strings => set relative to current level


say.set(indent=0)      # to get back to the default, no indent

Or you can use a with construct:

with say.settings(indent='+1'):

    # anything say() emits here will be auto-indented +1 levels

# anything say() emits here, after the with, will not be indented +1


If using a string to indicate relative indent levels offends your sense of dimensionality or strict typing, there is a class Relative that does the same thing in a more formal way. indent='+2' and indent=Relative(+2) are identical.

If you have a lot of data or text to print, and it would normally create super-long, difficult-to-read lines, you can easily wrap it:

say("This is a really long...blah blah blah", wrap=40)

Will automatically wrap the text to the given width using Python’s standard textwrap module. Feel free to use indentation and wrapping together.

Prefixes and Suffixes

Every line can be given a prefix or suffix, if desired. For example:

with say.settings(prefix='> '):

Will give what text email and Markdown consider a quoted block look:

> this
> that

Or if you’d like some text to be quoted with blue quotes:

say(text, prefix=styled('> ', 'blue'))

And if you like your output numbered:



1: this
2: and
3: that

You can instantiate different numberers for different files, and if you like, use the start keyword argument to start a numberer on a designated value.

The Value Proposition

While it’s easy enough to add a few spaces to the format string of any print statement or function in order to achieve a little indentation, it’s easy to mistakenly type too many or too few spaces, or to forget to type them in some format strings. If you’re indenting strings that themselves may contain multiple lines, the simple print approach breaks because it won’t take multi-line strings into account. Nor will it be integrated with line wrapping or numbering or other formatting you also want.

say, however, simply and correctly handles these combined formatting operations. Harder cases like multi-line strings are just as nicely and well indented as simple ones–something not otherwise easily accomplished without adding gunky, complexifying string manipulation code to every place in your program that prints anything.

This starts to illustrate say‘s “do the right thing” philosophy. So many languages’ printing and formatting functions “output values” at a low level. They may format basic data types, but they don’t provide straightforward ways to do neat text transformations that rapidly yield correct, attractively-formatted output. say does. Over time, say will provide even more high-level formatting options. For now: indentation, wrapping, and line numbering.


If you do find any errors in the way say handles formatting operations, there’s an app for that. Let’s fix them once, in a common place, in reusable code–not spread around many different programs.

Titles and Horizontal Rules

say defines a few convenience formatting functions:

say.title('Errors', sep='-')
for i,e in enumerate(errors, start=1):
    say("{i:3}: {e['name'].upper()}")

might yield:

--------------- Errors ---------------
  1: I/O ERROR

A similar method hr produces just a horizontal line (“rule”), like the HTML <hr> element. For either, one can optionally specify the width (width), character repeated to make the line (sep), and vertical separation/whitespace above and below the item (vsep). Good options for the separator might be be ‘-‘, ‘=’, or parts of the Unicode box drawing character set.

Vertical Separation

You don’t need to add newline characters here and there to achieve good vertical spacing. say.blank_lines(n) emits n blank lines. And just about every say call also supports a vsep (vertical separation) parameter.:

say('TITLE', vsep=(2,0)        # add 2 newlines before (none after)
say('=====', vsep=(0,2))       # add 2 newlines after (none before)
say('something else', vsep=1)  # add 1 newline before, 1 after

Colors and Styles

say has built-in support for style-driven formatting. By default, ANSI terminal colors and styles are automagically supported.

answer = 42

say("The answer is {answer:style=bold+red}")

This uses the ansicolors module, though with a slightly more permissive syntax. Available colors are ‘black’, ‘blue’, ‘cyan’, ‘green’, ‘magenta’, ‘red’, ‘white’, and ‘yellow’. Available styles are ‘bold’, ‘italic’, ‘underline’, ‘blink’, ‘blink2’, ‘faint’, ‘negative’, ‘concealed’, and ‘crossed’. These styles can be combined with a + or | character. Note, however, that not all styles are available on every terminal.


When naming a style within the template braces ({}) of format strings, you can quote the style name or not. fmt("{x:style=red+bold}") is equivalent to fmt("{x:style='red+bold'}").

You can define your own styles: x: color(x, fg='red'))

Because styles are defined through executables (lambdas, usually), they can include decisions or text transformations of arbitrary complexity. For example: n: color(n, fg='red', style='bold') if int(n) < 0 else n)
say("Result: {n:style=redwarn}")

That will display the number n in bold red characters, but only if it’s value is negative. For positive numbers, n is displayed normally.

Or define a style where a message is surrounded by red stars: x: fmt('*** ', style='red') + \
                          fmt(x,      style='black') + \
                          fmt(' ***', style='red')) x: 'x' * len(x))

message = 'hey'
say(message, style='stars')
say(message, style='redacted')


*** hey ***

(with red stars)


Style defining lambdas (or functions) take string arguments. If the string is logically a number, it must be then cast into an int, float, or whatever. The code must ultimate return a string.

You can also apply a style to the entire contents of a say or fmt invocation:

say("There is green everywhere!", style='green|underline')

While the goal of say is to have correct behavior under absolutely all combinations of text styling, coloring, indentation, numbering, and so on, be aware that the coloring/styling is relatively new, has limited tests and documentation to date, and is still evolving. One known bug attends say‘s use of Python’s textwrap module, which is not savvy to ANSI-terminal control codes; text that includes control codes and is wrapped is currently likely to wrap in the wrong place. Enclosing one bit of colored text inside another bit of colored text is not as easy as it could be. Finally, style definitions are idiosyncratically shared across instances. That said, some fairly complex invocations already work quite nicely. Try, e.g.:

say.set(prefix=numberer(template=color('{n:>3}: ', fg='blue')), \
say('a long paragraph...with gobs of text', style='red')

This correctly puts the line numbers in blue, wraps the lines to 40 characters, and puts the text in red. (The textwrap collision with control characters is avoided here because the wrapped text is pure, and the control codes for red styling are added after wrapping.)

Styled formatting is an extremely powerful approach, giving the same kind of flexibility and abstraction seen for styles in word processors and CSS-based Web design. It will be further developed. Plans already include replacing textwrap with an ANSI-savvy text wrapping module, providing simpler ways to state complex formatting, and mechanisms to auto-map styles into HTML output.

Where You Like

say() writes to a list of files–by default just sys.stdout. But with one simple configuration call, it will write to different–even multiple–files:

import sys
from say import say

say.set(files=[sys.stdout, "report.txt"])
say(...)   # now prints to both sys.stdout and report.txt

This has the advantage of allowing you to both capture and see program output, without changing any code (other than the config statement). You can also define your own targeted Say instances:

import sys
from say import say

err = say.clone(files=[sys.stderr, 'error.txt'])
err("Failed with error {errcode}")  # writes in both places
say("something else")   # independent of err

When You Like

If you want to stop printing for a while:

say.set(silent=True)  # no printing until set to False

Or transiently:

say(...stuff..., silent=not verbose) # prints iff bool(verbose) is True

Of course, you don’t have to print at all. fmt() works exactly like say() and inherits most of its options, but doesn’t print. (The C analogy: say : fmt :: printf : sprintf.)


Character encodings remain a fractious and often exasperating part of IT. say() and fmt() try to avoid this by working with Unicode strings. In Python 3, all strings are Unicode strings, and output is by default UTF-8 encoded. Yay!

In Python 2, we try to maintain the same environment. If a template or input string is not of type unicode, please include only ASCII characters, not encoded bytes from UTF-8 or whatever. If you don’t do this, any trouble results be on your head. If say opens a file for you (e.g. with setfiles()), it uses to inherit its default encoding to UTF-8. If you have say write to a file that you’ve opened, you should similarly use or another mechanism that transparently writes to a proper encoding.

Non-Functional Invocation

For those who don’t want to always and forever surround “print statements” with the Python 3-style function parentheses, the > operator is provided as an experimental, non-functional way to print. The following are identical:

say> "{}: {user.username}"
say("{}: {user.username}")

You can name as many values as you like in the format string, but there can only be one format string, and no options. If you need to say multiple values, or say them with statement-specific options, you must use the functional syntax.

Text and Templates

Often the job of output is not about individual text lines, but about creating multi-line files such as scripts and reports. This often leads away from standard output mechanisms toward template packages, but say has you covered here as well.

from say import Text

# assume `hostname` and `filepath` already defined

script = Text()
script += """

    # Output the results of a ping command to the given file

    ping {hostname!r} >{filepath!r}


Then will contain:


# Output the results of a ping command to the given file

ping '' >'ping-results.txt'

Text objects are basically a list of text lines. In most cases, when you add text (either as multi-line strings or lists of strings), Text will automatically interpolate variables the same way say does. One can simply print or say Text objects, as their str() value is the full text you would assume. Text objects have both text and lines properties which can be either accessed or assigned to.

+= incremental assignment automatically removes blank starting and ending lines, and any whitespace prefix that is common to all of the lines (i.e. it will dedent any given text). This ensures you don’t need to give up nice Python program formatting just to include a template.

While += is a handy way of incrementally building text, it isn’t strictly necessary in the simple example above; the Text(...) constructor itself accepts a string or set of lines.

Other in-place operators are: |= for adding text while preserving leading white space (no dedent) and &= adds text verbatim–without dedent or string interpolation.

One can read_from() a file (appending the contents of the file to the given text object, with optional interpolation and dedenting). One can also write_to() a file. Use the append flag if you wish to add to rather than overwrite the file of a given name, and you can set an output encoding if you like (encoding='utf-8' is the default).

So far we’ve discussed``Text`` objects almost like strings, but they also act as lists of individual lines (strings). They are, for example, indexable via [], and they are iterable. Their len() is the number of lines they contain. One can append() or extend() them with one or multiple strings, respectively. append() takes a keyword parameter interpolate that controls whether {} expressions in the string are interpolated. extend() additionally takes a dedent flag that, if true, will automatically remove blank starting and ending lines, and any whitespace prefix that is common to all of the lines.

If t is a Text instance, str(t) will be the full string representing it. If you wish to move from multiple lines to a single-line, joined string, ' '.join(t) does the trick.

Text objects, unlike strings, are mutable. The replace(x, y) method will replace all instances of x with y in situ. If given just one argument, a dict, all the keys will be replaced with their corresponding values.

Text doesn’t have the full set of text-onboarding options seen in textdata, but it should suit many circumstances. If you need more, textdata can be used alongside Text.

Finally, it’s possible to use a Text object like a file and write to it. So:

t = Text()
say.set(files=[sys.stdout, t])


will now append each thing said to both sys.stdout and t.

There is a related class Template that does not interpolate its format variables when constructed, but rather when explicitly rendered. This suits certain form-filling operations:

t = Template("Dear {name},\n\nWelcome to our club!\n")
for name in 'Joe Jane Jeremey'.split():
    print t.render()

Iterpolators and Exceptions

You may want to write your own functions that take strings and interpolate {} format templates in them. The easy way is:

from say import caller_fmt

def ucfmt(s):
    return caller_fmt(s).upper()

If ucfmt() had used fmt(), it would not have worked. fmt() would look for interpolating values within the context of ucfmt() and, not finding any, probably raised an exception. But using caller_fmt() it looks into the context of the caller of ucfmt(), which is exactly where those values would reside. Voila!

And example of how this can work–and a useful tool in its own right–is FmtException. If you want to have comprehensible error messages when something goes wrong, you could use fmt():

if bad_thing_has_happened:
    raise ValueError(fmt("Parameters {x!r} or {y!r} invalid."))

But if you define your own exceptions, consider subclassing FmtException:

class InvalidParameters(FmtException, ValueError):


if bad_thing_has_happened:
    raise InvalidParameters("Parameters {x!r} or {y!r} invalid.")

You’ll save a few characters, and the code will be simpler and more comprehensible.

Python 3

Say works virtually the same way in Python 2 and Python 3. This can simplify software that should work across the versions, without the hassle of from __future__ import print_function.

say attempts to mask some of the quirky complexities of the 2-to-3 divide, such as string encodings and codec use.


  • ScopeFormatter provides variable interpolation into strings. It is amazingly compact and elegant. Sadly, it only interpolates Python names, not full expressions. say has full expressions, as well as a framework for higher-level printing features beyond ScopeFormatter‘

  • interpolate is similar to say.fmt(), in that it can interpolate complex Python expressions, not just names. It’s i % "format string" syntax is a little odd, however, in the way that it re-purposes Python’s earlier "C format string" % (values) style % operator. It also depends on the native print statement or function, which doesn’t help bridge Python 2 and 3.

  • Even simpler are invocations of % or format() using locals(). E.g.:

    name = "Joe"
    print "Hello, %(name)!" % locals()
    # or
    print "Hello, {name}!".format(**locals())

    Unfortunately this has even more limitations than ScopeFormatter: it only supports local variables, not globals or expressions. And the interpolation code seems gratuitous. Simpler:

    say("Hello, {name}!")


  • The say name was inspired by Perl’s say, but the similarity stops there.
  • The show debug printing functions previously in this package have been split into a separate package, show.
  • A new text aggregation class, Text is now available.
  • Automated multi-version testing managed with the wonderful pytest and tox. Successfully packaged for, and tested against, all late-model versions of Python: 2.6, 2.7, 3.2, 3.3, and 3.4, as well as PyPy 2.6.0 (based on 2.7.9) and PyPy3 2.4.0 (based on 3.2.5). Should run fine on Python 3.5, though py.test is broken on its pre-release iterations.
  • say has greater ambitions than just simple template printing. It’s part of a larger rethinking of how output should be formatted. show <>`_ and Text are other down-payments on this larger vision. Stay tuned.
  • In addition to being a practical module in its own right, say is testbed for options, a package that provides high-flexibility option, configuration, and parameter management.
  • The author, Jonathan Eunice or @jeunice on Twitter welcomes your comments and suggestions. If you’re using say in your own work, drop me a note and tell me how you’re using it, how you like it, and what you’d like to see!


  • Use a text wrapping module that is fully cognient of ANSI escape codes.
  • Further formatting techniques for easily generating HTML output and formatting non-scalar values.
  • Provide code that allows pylint to see that variables used inside the say and fmt format strings are indeed thereby used.


To install or upgrade to the latest version:

pip install -U say

To easy_install under a specific Python version (3.3 in this example):

python3.3 -m easy_install --upgrade say

(You may need to prefix these with sudo comamnd to authorize installation. In environments without super-user privileges, you may want to use pip‘s --user option, to install only for a single user, rather than system-wide.)