advanced string formatting vs template strings

One key advantage of string templates is that you can substitute only some of the placeholders using the safe_substitute method. Normal format strings will raise an error if a placeholder is not passed a value. For example:

"Hello, {first} {last}".format(first='Joe')

raises:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
KeyError: 'last'

But:

from string import Template
Template("Hello, $first $last").safe_substitute(first='Joe')

Produces:

'Hello, Joe $last'

Note that the returned value is a string, not a Template; if you want to substitute the $last you'll need to create a new Template object from that string.

For what it's worth, Template substitution from a dict appears to be 4 to 10 times slower than format substitution, depending on the length of the template. Here's a quick comparison I ran under OS X on a 2.3 GHz core i7 with Python 3.5.

from string import Template
lorem = "Lorem ipsum dolor sit amet {GIBBERISH}, consectetur adipiscing elit {DRIVEL}. Expectoque quid ad id, quod quaerebam, respondeas."
loremtpl = Template("Lorem ipsum dolor sit amet $GIBBERISH, consectetur adipiscing elit $DRIVEL. Expectoque quid ad id, quod quaerebam, respondeas.")
d = dict(GIBBERISH='FOOBAR', DRIVEL = 'RAXOOP')

In [29]: timeit lorem.format(**d)
1.07 µs ± 2.13 ns per loop (mean ± std. dev. of 7 runs, 1000000 loops each)

In [30]: timeit loremtpl.substitute(d)
8.74 µs ± 12.9 ns per loop (mean ± std. dev. of 7 runs, 100000 loops each)

The worst case I tested was about 10 times slower for a 13 character string. The best case I tested was about 4 times slower for a 71000 character string.

Its primarily a matter of syntax preference, which usually boils down to a laziness/verbosity tradeoff and familiarity/habits with existing string template systems. In this case template strings are more lazy/simple/quick to write, while .format() is more verbose and feature-full.

Programmers used to the PHP language or the Jinja family of template systems may prefer template strings. Using "%s" positional style tuple substitution might appeal to those who use printf-like string formatting or want something quick. .format() has a few more features, but unless you need something specific that only .format() provides, there is nothing wrong with using any existing scheme.

The only thing to be aware of is that named string templates are more flexible and require less maintenance than order-dependent ones. Other than that it all comes down to either personal preference or the coding standard of the project you are working on;

This is old but it is worth mentioning that one great advantage of using Template Strings is that it is safe when you are accepting the template from un-trusted source. This can be from a configuration file for instance.

Here is an example from this article:

CONFIG = {"SECRET_KEY": "super secret key"}

class Event:
    def __init__(self, id_, level, message):
        self.id_ = id_
        self.level = level
        self.message = message

def format_event(format_string, event):
    return format_string.format(event=event)


# user supplied template
inp = "{event.__init__.__globals__[CONFIG][SECRET_KEY]}"

event = Event(1234, "foo", "boo")
print(format_event(inp, event))

output:

super secret key

Adding to the other answers: one advantage of template strings vs. format strings is that often, the characters {} are more likely to naturally occur in a piece of text than $foo or ${foo} (e.g. when generating source code). That means that with templates, you need to spend less effort in escaping these occurrences of proper text.