Quasiquotations for a python like interpolated string formatter

Version on this page:
LTS Haskell 22.29:
Stackage Nightly 2024-07-13:
Latest on Hackage:

See all snapshots PyF appears in

BSD-3-Clause licensed by Guillaume Bouchard
Maintained by [email protected]
This version can be pinned in stack with:PyF-,2791
Used by 1 package in lts-21.7(full list with versions):


PyF is a Haskell library for string interpolation and formatting.

PyF exposes a quasiquoter f which introduces string interpolation and formatting with a mini language inspired from printf and Python.

Quick Start

>>> import PyF

>>> name = "Dave"
>>> age = 54

>>> [fmt|Person's name is {name}, age is {age}|]
"Person's name is Dave, age is 54"

The formatting mini language can represent:

  • Numbers with different representations (fixed point, general representation, binary, hexadecimal, octal)
  • Padding, with the choice of padding char, alignment (left, right, around, between sign and number)
  • Sign handling, to display or not the + for positive number
  • Number grouping
  • Floating point representation
  • The interpolated value can be any Haskell expression.

You will need the extension QuasiQuotes, enable it with {-# LANGUAGE QuasiQuotes #-} in top of your source file or with :set -XQuasiQuotes in your ghci session. ExtendedDefaultRules and OverloadedStrings may be more convenient.

Expression to be formatted are referenced by {expression:formattingOptions} where formattingOptions follows the Python format mini-language. It is recommended to read the python documentation, but the Test file as well as this readme contain many examples.

More Examples


Left < / Right > / Around ^ padding:

>>> name = "Guillaume"
>>> [fmt|{name:<11}|]
"Guillaume  "
>>> [fmt|{name:>11}|]
"  Guillaume"
>>> [fmt|{name:|^13}|]

Padding inside = the sign

>>> [fmt|{-3:=6}|]
"-    3"

Float rounding

>>> [fmt|{pi:.2}|]

Binary / Octal / Hex representation (with or without prefix using #)

>>> v = 31
>>> [fmt|Binary: {v:#b}|]
"Binary: 0b11111"
>>> [fmt|Octal: {v:#o}|]
"Octal: 0o37"
>>> [fmt|Octal (no prefix): {v:o}|]
"Octal (no prefix): 37"
>>> [fmt|Hexa (caps and prefix): {v:#X}|]
"Hexa (caps and prefix): 0x1F"


Using , or _.

>>> [fmt|{10 ^ 9 - 1:,}|]
>>> [fmt|{2 ^ 32  -1:_b}|]

Sign handling

Using + to display the positive sign (if any) or to display a space instead:

>>> [fmt|{pi:+.3}|]
>>> [fmt|{-pi:+.3} (Negative number)|]
"-3.142 (Negative number)"
>>> [fmt|{pi: .3}|]
" 3.142"
>>> [fmt|{-pi: .3} (Negative number)|]
"-3.142 (Negative number)"


Preceding the width with a 0 enables sign-aware zero-padding, this is equivalent to inside = padding with a fill char of 0.

>>> [f{10:010}|]
>>> [f{-10:010}|]


First argument inside the curly braces can be a valid Haskell expression, for example:

>>> [fmt|2pi = {2* pi:.2}|]
2pi = 6.28
>>> [fmt|tail "hello" = {tail "hello":->6}|]
"tail \"hello\" = --ello"

However the expression must not contain } or : characters.


Most options can be combined. This generally leads to totally unreadable format string ;)

>>> [fmt|{pi:~>5.2}|]

Multi-line strings

You can ignore a line break with \ if needed. For example:

- a
- b\

Will returns -a\n-b. Note how the first and last line breaks are ignored.

Arbitrary value for precision and padding

The precision and padding width fields can be any Haskell expression (including variables) instead of a fixed number:

>>> [fmt|{pi:.{1+2}}|]
>>> [fmt|{1986:^{2 * 10}d}|]
"        1986        "

Output type

PyF aims at extending the string literal syntax. As such, it default to String type. However, if the OverloadedString is enabled, PyF will happilly generate IsString t => t instead. This means that you can use PyF to generate String, but also Text and why not ByteString, with all the caveats known to this extension.

>>> [fmt|hello {pi.2}|] :: String
"hello 3.14"

Custom types

PyF can format three categories of input types:

  • Floating. Using the f, g, e, … type specifiers. Any type instance of RealFloat can be formated as such.
  • Integral. Using the d, b, x, o, … type specifiers. Any type instance of Integral can be formated as such.
  • String. Using the s type specifier. Any type instance of PyFToString can be formated as such.

See PyF.Class if you want to create new instances for the PyFToString class.

By default, if you do not provide any type specifier, PyF uses the PyFClassify type class to decide if your type must be formated as a Floating, Integral or String.


Type inference

Type inference with numeric literals can be unreliable if your variables are too polymorphic. A type annotation or the extension ExtendedDefaultRules will help.

>>> v = 10 :: Double
>>> [fmt|A float: {v}|]
A float: 10

Error reporting

Template haskell is generally known to give developers a lot of frustration when it comes to error message, dumping an unreadable piece of generated code.

However, in PyF, we took great care to provide clear error reporting, this means that:

  • Any parsing error on the mini language results in a clear indication of the error, for example:
foo = [fmt|{age:.3d}|]
File.hs:77:19: error:
1 | {age:.3d}
  |        ^
Type incompatible with precision (.3), use any of {'e', 'E', 'f', 'F', 'g', 'G', 'n', 's', '%'} or remove the precision field.

Note: error reporting uses the native GHC error infrastructure, so they will correctly appear in your editor (using HLS), for example:

Error reported in editor

  • Error in variable name are also readable:
test/SpecUtils.hs:81:33: error:
    • Variable not found: chien
    • In the quasi-quotation: [fmt|A missing variable: {chien}|]
81 | fiz = [fmt|A missing variable: {chien}|]
   |                                 ^^^^^
  • However, if the interpolated name is not of a compatible type (or too polymorphic), you will get an awful error:
>>*> [fmt|{True:d}|]

<interactive>:80:10: error:
    • No instance for (Integral Bool)
        arising from a use of ‘PyF.Internal.QQ.formatAnyIntegral’
  • There is also one class of error related to alignement which can be triggered, when using alignement inside sign (i.e. =) with string:
*PyF PyF.Internal.QQ> [fmt|{"hello":=10}|]

<interactive>:89:10: error:
    • String Cannot be aligned with the inside `=` mode
  • Finally, if you make any type error inside the expression field, you are on your own, you’ll get an awful error in the middle of the generated template Haskell splice.
>>> [fmt|{3 + pi + "hello":10}|]

<interactive>:99:10: error:
    • No instance for (Floating [Char]) arising from a use of ‘pi’

Custom Delimiters

If { and } does not fit your needs, for example if you are formatting a lot of json, you can use custom delimiters. All quasi quoters have a parametric form which accepts custom delimiters. Due to template haskell stage restriction, you must define your custom quasi quoter in an other module.

For example, in MyCustomDelimiter.hs:

module MyCustomQQ where

import Language.Haskell.TH.Quote

import PyF

myCustomFormatter :: QuasiQuoter
myCustomFormatter = mkFormatter "fmtWithDelimiters" (fmtConfig {
  delimiters = ('@','!')

Later, in another module:

import MyCustomQQ

-- ...

[myCustomFormatter|pi = @pi:2.f!|]

Escaping still works by doubling the delimiters, @@!!@@!! will be formatted as @!@!.

Have a look at PyF.mkFormatter for all the details about customization.

Difference with the Python Syntax

The implementation is unit-tested against the reference python implementation (python 3.6.4) and should match its result. However some formatters are not supported or some (minor) differences can be observed.

Not supported

  • Number n formatter is not supported. In python this formatter can format a number and use current locale information for decimal part and thousand separator. There is no plan to support that because of the impure interface needed to read the locale.
  • Python support sub variables in the formatting options in every places, such as {expression:.{precision}}. We only support it for precision and width. This is more complexe to setup for others fields.
  • Python literal integers accepts binary/octal/hexa/decimal literals, PyF only accept decimal ones, I don’t have a plan to support that, if you really need to format a float with a number of digit provided as a binary constant, open an issue.
  • Python support adding custom formatters for new types, such as date. This may be really cool, for example [fmt|{today:%Y-%M-%D}. I don’t know how to support that now.


  • General formatters g and G behaves a bit differently. Precision influence the number of significant digits instead of the number of the magnitude at which the representation changes between fixed and exponential.
  • Grouping options allows grouping with an _ for floating point, python only allows ,.
  • Custom delimiters

Build / test

Should work with stack build; stack test, and with cabal and (optionally) nix:

nix-shell # Optional, if you use nix
cabal new-build
cabal new-test

There are a few available shells for you.

  • nix-shell is the default, current GHC version with language server available.
  • nix-shell ./. -A is another GHC version (change xx) without language server.
  • nix-shell ./. -A pyf_xx.shell_hls is another GHC version (change xx) with language server.

We also provide a few utility functions:

  • nix-build ./ -A hlint will check hlint.
  • nix-shell ./ -A ormolu-fix will format the codebase.

Library note

PyF.Formatters exposes two functions to format numbers. They are type-safe (as much as possible) and comes with a combination of formatting options not seen in other formatting libraries:

>>> formatIntegral Binary Plus (Just (20, AlignInside, '~')) (Just (4, ',')) 255

GHC compatibility

This library is tested in CI with ghc 8.6 to 9.2.


Don’t hesitate to make any suggestion, I’ll be more than happy to work on it.


Everything works with nix and flakes. But you can also try with manual cabal / stack if you wish.

  • nix develop will open a shell with everything you need to work on PyF, including haskell-language-server. It may be a bit too much, so you can instead:
  • nix develop .#pyf_XY opens a shell with a specific GHC version and without haskell-language-server. That’s mostly to test compatibility with different GHC version or open a shell without HLS if you are in a hurry. Replace pyf_XY by pyf_86, pyf_88, pyf_810, pyf_90 or pyf_92.

Once in the shell, use cabal build, cabal test, cabal repl.

There is a cachix available, used by CI, and already configured in flakes. You can manually run cachix use guibou if you want.

You can locally build and test everything using:

  • nix build .#pyf_all.

Don’t hesitate to submit a PR not tested on all GHC versions.


The codebase is formatted with ormolu. Please run:

  • nix run .\#run-ormolu

Before submitting.

Treesitter support

Have a look in the ./tree-sitter-pyf/ directory for a parser of PyF which can be integrated in your treesitter compatible editor to get syntax highlighting for PyF.


Revision history for PyF – 2023-03-15

  • Support for GHC 9.6. Thank you @Kleidukos for initiating the port. – 2022-09-24

  • Support for OverloadedRecordsDot syntax in Meta. Thank you @Profpatsch for the report.
  • In some context, the error reporting for variable not found in the quasi quote expression was incorrectly reporting existing variables as not found. See for details. This is now fixed by not abusing GHC api. Thank you @michaelpj for reporting this really weird problem. – 2022-08-10

  • Support for GHC 9.4. (Written with a pre-release of GHC 9.4, hopefully it won’t change too much before the release).
  • Error reporting now uses the native GHC API. In summary, it means that haskell-language-server will point to the correct location of the error, not the beginning of the quasi quotes.
  • PyF will now correctly locate the error for variable not found in expression, even if the expression is complicated. The support for complex expression is limited, and PyF may return a false positive if you try to format a complex lambda / case expression. Please open a ticket if you need that.
  • Add support for literal [] and () in haskell expression.
  • Add support for overloaded labels, thank you Shimuuar.
  • Support for :: in haskell expression. Such as [fmt| 10 :: Int:d}|], as a suggestion from julm (close #87).
  • Integral padding width and precision also for formatter without type specifier.
  • Extra care was used to catch all type-defaults warning message. PyF should not generate code with this kind of warning, unless the embedded Haskell expression are ambiguous (e.g. [fmt|{10}|]). You can use :: to disambiguate, e.g. [fmt|{10 :: Int}|]. – 2021-12-05

  • Padding width can now be any arbitrary Haskell expression, such as [fmt|hello pi = {pi:<{5 * 10}}|].
  • Precision (and now padding width) arbitrary expression can now be any Integral and it is not limited to Int anymore.
  • (Meta): type expression are now parsed and hence allowed inside arbitrary Haskell expression for padding width and precision. For example, [fmt|Hello {pi:.{3 :: Int}}|]. – 2021-10-30

  • Due to the dependencies refactor, PyF no have no dependencies other than the one packaged with GHC. The direct result is that PyF build time is reduced to 6s versus 4 minutes and 20s before.

  • Remove the dependency to megaparsec and replaces it by parsec. This should have minor impact on the error messages.

  • Huge Change. The parsing of embeded expression does not depend anymore on haskell-src-ext and haskell-src-meta and instead depends on the built-in ghc lib.

  • Added instances for (Lazy)ByteString to PyFClassify and PyFToString. ByteString can now be integrated into format string, and will be decoded as ascii.

  • Relax the constraint for floating point formatting from RealFrac to Real. As a result, a few new type can be formatted as floating point number. One drawback is that some Integral are Real too and hence it is not an error anymore to format an integral as floating point, but you still need to explicitly select a floating point formatter.

  • Added instance for (Nominal)DiffTime to PyFClassify, so you can now format them without conversion.

  • Introducing of the new typeclass PyfFormatIntegral and PyfFormatFractional in order to customize the formatting for numbers. An instance is derived for respectively any Integral and Real types.

  • Support for Char formatting, as string (showing the Char value) or as integral, showing the ord.

  • Data.Ratio.

  • Introducing fmtTrim module. It offers the same behavior as fmt, but trims common indentation. Se PyF.trimIndent for documentation.

  • Introducing raw for convenience. It is a multiline string without any escaping, formatting neither leading whitespace handling.

  • Introducing str and strTrim. They are similar to fmt and fmtTrim but without formatting. You can see them as multiline haskell string, with special character escaping, but without formatting. For convenience, the strTrim version also removes indentation.

  • fmtWithDelimiters is gone and replaced by mkFormatter in PyF which is “more” generic. – 2021-02-06

  • Test phase do not depend anymore on python (unless cabal flag python_test is set). This ease the deployment / packaging process. – 2020-09-11

  • Version bump for megaparsec 9.0 – 2020-03-25

  • Fixs for GHC 8.10 – 2019-12-29

  • Any type with Show instance can be formatted using :s formatter. For example, [fmt|hello {(True, 10):s}|]. This breaks compatibility because previous version of PyF was generating an error when try to format to string anything which was not a string, now it accepts roughly anything (with a Show instance). – 2019-11-08

  • Bump megaparsec bounds – 2019-10-13

  • Compatibility with GHC 8.8 – 2019-09-03

  • Precision can now be any arbitrary haskell expression, such as [fmt|hello pi = {pi:.{1 + 3}}|]. – 2019-08-27

  • (minor bugfix in tests): Use python3 instead of “python” to help build on environment with both python2 and python3 – 2019-08-27

  • Stack support – 2019-08-06

  • f (and fWithDelimiters) were renamed fmt (fmtWithDelimiters). f was causing too much shadowing in any codebase.
  • PyF now exposes the typeclass PyFToString and PyFClassify which can be extended to support any type as input for the formatters.
  • PyF now uses Data.String.IsString t as its output type if OverloadedString is enabled. It means that it behaves as a real haskell string literal.
  • A caveat of the previous change is that PyF does not have instances for IO anymore.

bugfixes and general improvements

  • An important amount of bugfixs
  • Error reporting for generic formatting (i.e. formatting without a specified type) is now more robust
  • Template haskell splices are simpler. This leads to more efficient / small generated code and in the event of this code appears in a GHC error message, it is more readable.
  • PyF now longer emit unnecessary default typing. – 2019-02-28

  • Tests: fix non reproducible tests – 2019-02-27

  • Fixed: PyF now uses the same haskell extensions as the one used by the current haskell file to parse sub expressions. – 2019-02-11

  • Fixed: PyF was wrongly ignoring everything located after a non-doubled closing delimiter.
  • New Feature: line break can be escaped with , thus allowing string to start on a new line ignoring the initial backspace – 2019-02-04

  • Bump dependencies to megaparsec 7
  • Error message are now tested
  • Name in template haskell splices are stable. This increases readability of error messages
  • Better error message for badly formated expression

Formatting removal

  • All monomorphic quasiquoters (f, fString, fText, fIO, fLazyText) are removed
  • Polymophic quasiquoter f' is renamed f and is the only entry point. Monomorphic users are encouraged to use the polymorphic quasiquoter with type annotation.
  • Formatting dependency is removed.
  • Previously named f quasiquoters which was exporting to Formatting.Format is removed. User of this behavior should use instead. – 2018-08-03

  • Custom delimiters, you can use whatever delimiters you want in place of { and }. – 2018-08-02

  • Fix the espace parsing of {{ and }} as { and } – 2018-04-16

  • Support for negative zero
  • Support for 0 modifier
  • Exponential formatter now behaves as python
  • Support for alternate floatting point represenation
  • Lot of documentation
  • Test are auto verified with the python reference implementation – 2018-04-13

  • Support for grouping option
  • Support for inner allignment
  • Correct display of NaN and Infinity
  • Fix a few cosmetic with python implementation
  • Introduce PyF.Formatters, type safe generic number formatter solution
  • Remove dependency to scientific – 2018-04-01

  • Support for haskell subexpression – 2018-01-07

  • Add support for the sign field. – 2018-01-03

  • First version. Released on an unsuspecting world.