Complete the ORM Function Library

[Matthew Pava]

I was surprised to learn that we didn't have a StrIndex function until version 2, and yet we had Substr since at least version 1.8. I wonder how users were using Substr without also finding a use for StrIndex this whole time. Anyway, since we seem to be adding these functions one at a time, why don't we work on trying to get the built-ins implemented in one sweep instead?

We may even want to split the documentation page (​https://docs.djangoproject.com/en/dev/ref/models/database-functions/) into further categories with String functions and Numeric functions.

This is just a sample checklist, with corresponding attributes to which backend has them available.

Comparison
Django	SQL	Description	PostgreSQL	Oracle	MySQL	SQLite	PR
NullIf	NULLIF	Returns NULL if the first argument equals the second.	‎✔	‎✔	‎✔	‎✔	9543
Math
Django	SQL	Description	PostgreSQL	Oracle	MySQL	SQLite	PR
Abs	ABS	Returns the absolute value.	‎✔	‎✔	‎✔	‎✔	9622
ACos	ACOS	Returns the arccosine.	‎✔	‎✔	‎✔	✔1	9622
ASin	ASIN	Returns the arcsine.	‎✔	‎✔	‎✔	✔1	9622
ATan	ATAN	Returns the arctangent.	‎✔	‎✔	‎✔	✔1	9622
ATan2	ATAN2	Returns the arctangent of the two variables passed to it.	‎✔	‎✔	‎✔	✔1	9622
Ceil	CEILING	Returns the smallest integer value that is not less than a numeric expression	‎✔	CEIL	‎✔	✔1	9622
Cos	COS	Returns the cosine expressed in radians.	‎✔	‎✔	‎✔	✔1	9622
Cot	COT	Returns the cotangent.	‎✔	‎✔5	‎✔	✔1	9622
Degrees	DEGREES	Returns a numeric expression converted from radians to degrees.	‎✔	‎✔	‎✔	✔1	9622
Exp	EXP	Returns the base of the natural logarithm (e) raised to the power of a numeric expression.	‎✔	‎✔	‎✔	✔1	9622
Floor	FLOOR	Returns the largest integer value that is not greater than a numeric expression.	‎✔	‎✔	‎✔	✔1	9622
Ln	LN	Returns the natural logarithm of a numeric expression.	‎✔	‎✔	‎✔	✔1	9622
Log	LOG(B, X)	Returns the logarithm of a numeric expression.	‎✔	‎✔	‎✔	✔1	9622
Mod	MOD	Returns the remainder of one expression by diving by another expression.	‎✔	‎✔	‎✔	✔1	9622
Pi	PI	Returns the value of π	‎✔	✔4	‎✔	✔1	9622
Power	POWER	Returns the value of one expression raised to the power of another expression	‎✔	‎✔	‎✔	✔1	9622
Radians	RADIANS	Returns the value of an expression converted from degrees to radians.	‎✔	‎✔	‎✔	✔1	9622
Round	ROUND	Returns a numeric expression rounded to an integer. Can be used to round an expression to a number of decimal points	‎✔	‎✔	‎✔	‎✔	9622
Sin	SIN	Returns the sine given in radians.	‎✔	‎✔	‎✔	✔1	9622
Sqrt	SQRT	Returns the square root.	‎✔	‎✔	‎✔	✔1	9622
Tan	TAN	Returns the tangent expressed in radians.	‎✔	‎✔	‎✔	✔1	9622
Text
Django	SQL	Description	PostgreSQL	Oracle	MySQL	SQLite	PR
Ord	ASCII	Returns numeric value of left-most character. (Equivalent to ord() in Python.)	‎✔	‎✔	‎✔	UNICODE	9583
Chr	CHR	Character with the given code	‎✔	‎✔	CHAR	CHAR	9583
Left	LEFT	Returns the leftmost number of characters as specified	‎✔	✔2, 6	‎✔	✔2	9583
LPad	LPAD	Returns the string argument, left-padded with the specified string	‎✔	‎✔	‎✔	✔1	9798
LTrim	LTRIM	Removes leading spaces	‎✔	‎✔	‎✔	‎✔	9220
MD5	MD5	Calculates the MD5 hash of string, returning the result in hexadecimal	‎✔	✔8	✔	✔1	11004
Repeat	REPEAT	Repeats a string the specified number of times	‎✔	✔3	‎✔	✔1	9808
Replace	REPLACE	Replaces occurrences of a specified string	‎✔	‎✔6	‎✔	‎✔	9539
Reverse	REVERSE	Reverse the characters in a string	‎✔	✔6, 7	‎✔	✔1	10827
Right	RIGHT	Returns the specified rightmost number of characters	‎✔	✔2	‎✔	✔2	9583
RPad	RPAD	Appends string the specified number of times	‎✔	‎✔6	✔	✔1	9798
RTrim	RTRIM	Removes trailing spaces	‎✔	‎✔	‎✔	‎✔	9220
Trim	TRIM	Removes leading and trailing spaces	‎✔	‎✔	‎✔	‎✔	9220

• ¹ Function can be easily supported on SQLite with a ​user defined function.
• ² Can be emulated by using SUBSTR.
• ³ Can be emulated by using LENGTH and RPAD.
• ⁴ Can be emulated by directly substituting constant math.pi.
• ⁵ Can be emulated by using 1 / TAN(X).
• ⁶ Doesn't work properly with multibyte characters sets on Oracle.
• ⁷ The REVERSE function is undocumented on Oracle.
• ⁸ Can be supported by using STANDARD_HASH on Oracle.

[Josh Smeaton]

I had hoped that the 3rd party community would take care of providing "Function Packs" for each of the backends. But functions are fairly easy to create in your own project if you need them so I guess few people have bothered to group them into a distributable package.

I would prefer that the django ecosystem had a standard group of function expressions that can be used across backends, rather than a bunch of similar but not quite the same implementations. Then 3rd party libraries can depend on the known good versions and everyone is better off. Since the community hasn't seemed to provide such a library (that I'm aware of), I think it's fine for Django to do so.

Splitting the module up into string/numeric/date types sounds fine. Let's take inspiration from postgres and other db vendor docs so navigating them is familiar. Let's begin with functions that have standard support across our 4 backends, then look into what's left over. If we can approximate support by combining other functions we can do that. Otherwise if a particular function only has support for one or two backends we can consider ignoring it or implementing in a contrib module.

[Josh Smeaton]

For whoever wants to begin here, I think a good plan of attack will be a PR per function or per a small set of functions. Then the commit message would be something like "Refs #28643 -- Added X, Y, Z functions".

That way we can get through smaller pieces at a time, without such a large burden being placed on any one individual. We could also get some newer contributors to help in this way. Each function on its own should be a relatively easy thing to implement. A higher level task of re-organising the layout into functions/string functions/number may be helpful to get the ticket rolling.

[Matthew Pava]

I did find this third-party utility that handles specific PostgreSQL functions. I wonder if there are other backends that they could be ported to.
​https://github.com/hypertrack/django-pg-utils

And I also wonder how all of this connects with specific PostgreSQL aggregate functions already builtin to Django:
​https://docs.djangoproject.com/en/1.11/ref/contrib/postgres/aggregates/

[Mads Jensen]

Replying to Matthew Pava:

I did find this third-party utility that handles specific PostgreSQL functions. I wonder if there are other backends that they could be ported to.
​https://github.com/hypertrack/django-pg-utils

From a quick look, these things are already supported by Django.

[Mariusz Felisiak]

I prepared patch to reorganize database functions docs and code (​PR). I think we should organize code and doc as follows (bolded functions doesn't exist):

• docs/ref/models/database-functions.txt section Comparison and conversion functions (django/db/models/functions/comparison.py):
  • Cast, Coalesce, NullIf, Greatest, Least;
• docs/ref/models/database-functions.txt section Date Functions (django/db/models/functions/datetime.py):
  • Extract, ExtractDay, ExtractHour, ExtractMinute, ExtractMonth, ExtractQuarter, ExtractSecond, ExtractWeek, ExtractWeekDay, ExtractYear, Now, Trunc, TruncDate, TruncDay, TruncHour, TruncMinute, TruncMonth, TruncQuarter, TruncSecond, TruncTime, TruncYear";
• docs/ref/models/database-functions.txt section Math Functions (django/db/models/functions/math.py):
  • Abs, ACos, ASin, ATan, ATan2, Ceil, Cos, Cot, Exp, Floor, Log, Mod, Power, Round, Sin, Sqrt, Tan;
• docs/ref/models/database-functions.txt section Text Functions (django/db/models/functions/text.py):
  • Ascii, Chr, Concat, ConcatPair, Length, LPad, Lower, LTrim, Replace, RPad, RTrim, StrIndex, Substr, Trim, Upper;
• docs/ref/models/database-functions.txt section Window Functions (django/db/models/functions/window.py):
  • CumeDist, DenseRank, FirstValue, Lag, LastValue, Lead, NthValue, Ntile, PercentRank, Rank, RowNumber.

Headers and functions in docs and code should be organized alphabetically. Thanks Tim Graham for suggestions.

I'm not convince that following functions should be implemented because there not supported on all databases:

• LEFT(), MD5(), REPEAT(), REVERSE(), RIGHT(), DEGREES(), PI(), RADIANS().

[Tim Graham]

The "Miscellaneous" group is sort of bothering me.

Cast, Coalesce, Greatest, Least sound like candidates for a section called something like "Comparison and conversion functions"

Ascii and Chr sound like they could be "text" related.

Now sounds like a candidate for "Date Functions".

[Mariusz Felisiak]

Agreed. I updated above comment and both PRs.

[Matthew Pava]

In regards to Left() and Right(), they are available in PostgreSQL and MySQL, but not in SQLite or Oracle. However, you can emulate their functionality using the corresponding Substr function in the database backend. I would hate to see us leave those out for that reason alone. At the same time, developers could use the Substr across all databases instead of using Left and Right.

I was also wondering about the use of Trim. Python doesn't use that term; instead, it uses strip. I wonder if that really matters. Some developers may look for a strip functionality in Django ORM not realizing that the databases use the term Trim. Perhaps that could be clarified in the documentation.

[Shai Berger]

Replying to Matthew Pava:

I was also wondering about the use of Trim. Python doesn't use that term; instead, it uses strip. I wonder if that really matters. Some developers may look for a strip functionality in Django ORM not realizing that the databases use the term Trim. Perhaps that could be clarified in the documentation.

Unless there is some other use for the term strip in the databases, I'd consider "documenting" this by giving an alias:

class Trim(Transform):
    # ...

Strip = Trim  # Give Trim a more Pythonic name

[GitHub <noreply@…>]

In ad8036d7:

Refs #28643 -- Reorganized database functions docs.

Thanks Tim Graham for the review.

[Mariusz Felisiak <felisiak.mariusz@…>]

In 8b42a18b:

[2.0.x] Refs #28643 -- Reorganized database functions docs.

Thanks Tim Graham for the review.

Backport of ad8036d715d4447b95d485332511b4edb1a40c0e from master

[GitHub <noreply@…>]

In 4f27e475:

Refs #28643 -- Reorganized database functions.

Thanks Tim Graham for the review.

[Mariusz Felisiak]

​PR - Ltrim, Rtrim, and Trim with aliases Lstrip, Rstrip, and Strip, respectively.

[Matthew Pava]

And, of course, there is a Python lstrip and rstrip as well.

Lstrip = Ltrim  # Give Ltrim a more Pythonic name
Rstrip = Rtrim  # Give Rtrim a more Pythonic name

[Mariusz Felisiak]

Replying to Matthew Pava:

And, of course, there is a Python lstrip and rstrip as well.

Lstrip = Ltrim  # Give Ltrim a more Pythonic name
Rstrip = Rtrim  # Give Rtrim a more Pythonic name

I added these aliases.

[Matthew Pava]

I formatted the list into a table so we could see better what we can do to address this ticket. It looks like SQLite is missing much functionality that the other backends support, especially in the math department.

I also suggest implementing Left and Right, but for the unsupported backends to use the corresponding Substr function.

[Nick Pope]

Updated description and comment:9 to include NullIf.

See the following links:

• ​http://modern-sql.com/concept/null#nullif
• ​http://modern-sql.com/feature/case#nullif
• ​http://modern-sql.com/feature/case#compatibility

[Mads Jensen]

Added link to NullIf PR.

[Thomas Lagae]

Add link to Abs PR.

I am a first time contributor so any feedback is much appreciated.

[Thomas Lagae]

When passing a non-numeric value into the Abs function, the behaviour differs between databases. I think this will also be the case for other math functions. Should we catch non-numeric values and throw an error or just document that this behaviour is undefined and differs between databases?

[Matthew Pava]

I added a link to the Ord, Chr, Left, and Right PR.

I just started thinking more about the Left and Right aspects, and it occurred to me that a more pythonic solution would be the implementation of slicing instead of having these separate database functions. This would also change the implementation of Substr since that would no longer be necessary either.

[Tim Graham <timograham@…>]

In 65728550:

Refs #28643 -- Added Replace database function.

[Tim Graham <timograham@…>]

In f82de6bf:

Refs #28643 -- Added Ord, Chr, Left, and Right database functions.

[GitHub <noreply@…>]

In 9421aee:

Refs #28643 -- Added LTrim, RTrim, and Trim database functions.

Thanks Tim Graham and Mads Jensen for reviews.

[GitHub <noreply@…>]

In cede5111:

Refs #28643 -- Added LPad and RPad database functions.

Thanks Tim Graham for the review.

[GitHub <noreply@…>]

In 55cc269:

Refs #28643 -- Added Repeat database function.

Thanks Tim Graham and Nick Pope for reviews.

[Tim Graham <timograham@…>]

In a0b19a0:

Refs #28643 -- Added math database functions.

Thanks Nick Pope for much review.

[Tim Graham <timograham@…>]

In c3bbf1fd:

Refs #28643 -- Skipped ATan2() workaround on SpatiaLite 5+.

[Nick Pope]

I have added a new ​PR that cleans up some of the compatibility mixins added as part of the math function work so that they can be reused.
It also makes use of these mixins for aggregates defined in django.db.models.aggregates.

[Nick Pope]

Added a ​PR implementing Reverse().
Note that REVERSE exists for Oracle but is undocumented and works on bytes, not multi-byte strings. (Support table updated in ticket description.)

[Tim Graham <timograham@…>]

In 4b9d7221:

Refs #28643 -- Added NullIf database function.

Thanks Nick Pope, Mariusz Felisiak, and Tim Graham for reviews.

[Tim Graham <timograham@…>]

In abf8e390:

Refs #28643 -- Added Reverse database function.

Thanks Mariusz Felisiak for Oracle advice and review.

[Tim Graham <timograham@…>]

In 3d5e0f83:

Refs #28643 -- Moved db function mixins to a separate module.

[Tim Graham <timograham@…>]

In c690afb8:

Refs #28643 -- Changed Avg() to use NumericOutputFieldMixin.

Keeps precision instead of forcing DecimalField to FloatField.

[Tim Graham <timograham@…>]

In e85afa59:

Refs #28643 -- Changed StdDev() to use NumericOutputFieldMixin.

Keeps precision instead of forcing DecimalField to FloatField.

[Tim Graham <timograham@…>]

In 6d4efa8e:

Refs #28643 -- Changed Variance() to use NumericOutputFieldMixin.

Keeps precision instead of forcing DecimalField to FloatField.

[Tim Graham <timograham@…>]

In 846624ed:

Refs #28643 -- Extracted DurationField logic for Avg() and Sum() into mixin.

Also addresses Sum() not handling the filter option correctly.

[Mariusz Felisiak]

​PR implementing MD5().

[Nick Pope]

​PR implementing SHA1(), SHA224(), SHA256(), SHA384() and SHA512(). Moved to #30240.