rosa/tragedy

Watch

Fork 0

chore(deps): update dependency decimal to ~> 3.2.0 #7

Merged

rosa merged 1 commit from renovate/decimal-3.x into main

2026-10-17 20:11:36 +00:00

rosa commented

2026-10-17 20:11:34 +00:00

Owner

This PR contains the following updates:

Package	Type	Update	Change
decimal (source)	prod	minor	`~> 3.1.0` → `~> 3.2.0`

Release Notes

ericmj/decimal (decimal)

`v3.2.0`

Compare Source

Enhancements

Make most operations faster. Compared with v3.1.2 on OTP 29, on decimals
with up to 20 digits and exponents from -20 to 20, div/2 is about 8x
faster, mult/2 6x, add/2 5x, round/3 4x, compare/2 2.7x and
normalize/1 2x. On money amounts with two decimals, to_float/1 is about
23x faster, compare/3 18x, div/2 15x, rem/2 and sqrt/1 4x,
round/3 2-3x, add/2, sub/2, mult/2, div_int/2 and compare/2
2-2.7x, and parsing 1.4x. from_float/1 and cast/1 of a float are about
3.4x faster and cast/1 of an integer 3x. to_string/1 is unchanged.
Add the :subnormal and :clamped signals from the General Decimal
Arithmetic spec. :subnormal is signalled for every result whose adjusted
exponent is below emin, even an exact one, and :underflow only when
such a result is also inexact. A zero result's exponent is kept between
emin - precision + 1 and emax, the range of nonzero results, and
:clamped is signalled when it has to move:
Decimal.mult("0e-3500", "0e-3500") returns 0E-6176 instead of
0E-7000. This also applies to the zero results of division by infinity,
of div_int/2 and div_rem/2 with a smaller dividend, and of sqrt/1. A
subnormal result that rounds to zero signals :clamped too.

Bug fixes

Fix Decimal.div/2 rounding some inexact results the wrong way. Under
:half_even and :half_down, a result just above a tie was rounded as a
tie, and under :ceiling, :floor and :up, a result whose first
discarded digit was 0 was treated as exact even when later digits weren't,
so :ceiling and :floor could return a value on the wrong side of the
true quotient. About 5% of random divisions at the default precision were
affected. :inexact was also missing when the remainder was a power of
ten.
Fix :up rounding changing exact values. A result whose discarded digits
were all zero was still rounded away from zero: with precision: 1 and
rounding: :up, Decimal.add(0, Decimal.new("9.0")) returned 10
instead of 9, and Decimal.round(Decimal.new(0), -1, :up) returned
1E+1 instead of 0E+1.
Fix rounding carries leaving one digit more than the precision. Rounding
9.99 to precision 2 returned 10.0 instead of 10, and
Decimal.div(95, 10) at precision 1 returned 10 instead of 1E+1. This
affected every operation that rounds to the context precision, and the
results now match the General Decimal Arithmetic spec and Python's
decimal.
Fix Decimal.sqrt/1 rounding inexact roots as if the first discarded
digit were the whole remainder. At precision 9 under :ceiling,
Decimal.sqrt(10) returned 3.16227766 instead of 3.16227767, and at
precision 2 under :half_even, Decimal.sqrt("1.57") returned 1.2
instead of 1.3. The default :half_up rounding wasn't affected. Inexact
roots now signal :inexact as well as :rounded.
Fix Decimal.rem/2 and Decimal.div_rem/2 computing the remainder from a
product rounded to the context precision, which could cancel it out:
Decimal.rem("9999999999999999999999999999999999", "2.000000000000000000000000000000001")
returned 0 instead of 3E-33.
Fix Decimal.to_float/1 returning the next float up for odd integers
between 2^52 and 2^53, which doubles represent exactly:
Decimal.to_float(Decimal.new(4_503_599_627_370_497)) returned
4503599627370498.0.
Keep subnormal results instead of flushing them to zero. Since v3.0.0 set
the default emin to -6143, every result with an adjusted exponent below
it became 0 with :underflow, even an exact one:
Decimal.div("1e-6140", 10000) returned 0 instead of 1E-6144. Such
results are now rounded at the exponent emin - precision + 1 (-6176 by
default) with the context's rounding, as IEEE 754 and the General Decimal
Arithmetic spec specify, and signal :subnormal, plus :underflow when
that rounding is inexact. A result that rounds to zero keeps that exponent
(0E-6176). Decimal.round/3 no longer flushes an input below emin to
zero before rounding it.
Fix Decimal.round/3 rounding an input with more digits than the
precision twice, first with the context's rounding instead of mode: at
precision 3, Decimal.round(Decimal.new("0.4995"), 1, :down) returned
0.5 instead of 0.4.
Make Decimal.round/3 signal :rounded when it discards digits of a
nonzero coefficient, and :inexact when any of them is nonzero, as the
quantize operation of the General Decimal Arithmetic spec does.
Decimal.round("1.25", 1) returned 1.3 with no flags.
Make Decimal.round/3 signal :invalid_operation and return NaN where
quantize does: for ±Infinity, when -places is outside
emin - precision + 1 to emax, and when the result would need more
digits than the precision or an adjusted exponent above emax. The signal
is trapped by default, so these calls now raise Decimal.Error. Before,
the context rounded such results to an exponent other than -places:
Decimal.round(Decimal.new("1e40"), 2) returned
1.000000000000000000000000000000000E+40, and Decimal.round("1.5", 7000)
returned 1.500000000000000000000000000000000. ±Infinity was returned
unchanged.
Make Decimal.compare/3 exact. It rounded num1 ± threshold to the
context precision, so numbers near the threshold compared wrong:
Decimal.compare(1, "1.000000000000000000000000000000002", "1.5e-33")
returned :eq although they differ by 2e-33. It no longer sets
:inexact or :rounded, accepts -0 as a threshold, and raises
Decimal.Error instead of CondClauseError for a NaN when
:invalid_operation isn't trapped.
Make Decimal.compare/2 raise Decimal.Error for a NaN operand even when
:invalid_operation isn't trapped, instead of returning the NaN. Comparing
a NaN with ±Infinity now raises too, instead of returning :lt or :gt.
Make Decimal.Context.set/1, Decimal.Context.with/2 and
Decimal.Context.update/1 raise ArgumentError for an invalid context: a
precision that isn't a positive integer, an unknown rounding mode, an
emax or emin that is neither an integer nor :infinity, or an emin
above emax. A precision of 0 silently gave wrong results
(Decimal.add(1, 1) returned 0E+1), and the other cases failed with
FunctionClauseError in the next operation.
Make Decimal.to_integer/1 raise ArgumentError for NaN and ±Infinity,
like Decimal.to_float/1, instead of FunctionClauseError.

`v3.1.2`

Compare Source

Enhancements

Make most operations faster. Compared with v3.1.2 on OTP 29, on decimals
with up to 20 digits and exponents from -20 to 20, div/2 is about 8x
faster, mult/2 6x, add/2 5x, round/3 4x, compare/2 2.7x and
normalize/1 2x. On money amounts with two decimals, to_float/1 is about
23x faster, compare/3 18x, div/2 15x, rem/2 and sqrt/1 4x,
round/3 2-3x, add/2, sub/2, mult/2, div_int/2 and compare/2
2-2.7x, and parsing 1.4x. from_float/1 and cast/1 of a float are about
3.4x faster and cast/1 of an integer 3x. to_string/1 is unchanged.
Add the :subnormal and :clamped signals from the General Decimal
Arithmetic spec. :subnormal is signalled for every result whose adjusted
exponent is below emin, even an exact one, and :underflow only when
such a result is also inexact. A zero result's exponent is kept between
emin - precision + 1 and emax, the range of nonzero results, and
:clamped is signalled when it has to move:
Decimal.mult("0e-3500", "0e-3500") returns 0E-6176 instead of
0E-7000. This also applies to the zero results of division by infinity,
of div_int/2 and div_rem/2 with a smaller dividend, and of sqrt/1. A
subnormal result that rounds to zero signals :clamped too.

Bug fixes

Fix Decimal.div/2 rounding some inexact results the wrong way. Under
:half_even and :half_down, a result just above a tie was rounded as a
tie, and under :ceiling, :floor and :up, a result whose first
discarded digit was 0 was treated as exact even when later digits weren't,
so :ceiling and :floor could return a value on the wrong side of the
true quotient. About 5% of random divisions at the default precision were
affected. :inexact was also missing when the remainder was a power of
ten.
Fix :up rounding changing exact values. A result whose discarded digits
were all zero was still rounded away from zero: with precision: 1 and
rounding: :up, Decimal.add(0, Decimal.new("9.0")) returned 10
instead of 9, and Decimal.round(Decimal.new(0), -1, :up) returned
1E+1 instead of 0E+1.
Fix rounding carries leaving one digit more than the precision. Rounding
9.99 to precision 2 returned 10.0 instead of 10, and
Decimal.div(95, 10) at precision 1 returned 10 instead of 1E+1. This
affected every operation that rounds to the context precision, and the
results now match the General Decimal Arithmetic spec and Python's
decimal.
Fix Decimal.sqrt/1 rounding inexact roots as if the first discarded
digit were the whole remainder. At precision 9 under :ceiling,
Decimal.sqrt(10) returned 3.16227766 instead of 3.16227767, and at
precision 2 under :half_even, Decimal.sqrt("1.57") returned 1.2
instead of 1.3. The default :half_up rounding wasn't affected. Inexact
roots now signal :inexact as well as :rounded.
Fix Decimal.rem/2 and Decimal.div_rem/2 computing the remainder from a
product rounded to the context precision, which could cancel it out:
Decimal.rem("9999999999999999999999999999999999", "2.000000000000000000000000000000001")
returned 0 instead of 3E-33.
Fix Decimal.to_float/1 returning the next float up for odd integers
between 2^52 and 2^53, which doubles represent exactly:
Decimal.to_float(Decimal.new(4_503_599_627_370_497)) returned
4503599627370498.0.
Keep subnormal results instead of flushing them to zero. Since v3.0.0 set
the default emin to -6143, every result with an adjusted exponent below
it became 0 with :underflow, even an exact one:
Decimal.div("1e-6140", 10000) returned 0 instead of 1E-6144. Such
results are now rounded at the exponent emin - precision + 1 (-6176 by
default) with the context's rounding, as IEEE 754 and the General Decimal
Arithmetic spec specify, and signal :subnormal, plus :underflow when
that rounding is inexact. A result that rounds to zero keeps that exponent
(0E-6176). Decimal.round/3 no longer flushes an input below emin to
zero before rounding it.
Fix Decimal.round/3 rounding an input with more digits than the
precision twice, first with the context's rounding instead of mode: at
precision 3, Decimal.round(Decimal.new("0.4995"), 1, :down) returned
0.5 instead of 0.4.
Make Decimal.round/3 signal :rounded when it discards digits of a
nonzero coefficient, and :inexact when any of them is nonzero, as the
quantize operation of the General Decimal Arithmetic spec does.
Decimal.round("1.25", 1) returned 1.3 with no flags.
Make Decimal.round/3 signal :invalid_operation and return NaN where
quantize does: for ±Infinity, when -places is outside
emin - precision + 1 to emax, and when the result would need more
digits than the precision or an adjusted exponent above emax. The signal
is trapped by default, so these calls now raise Decimal.Error. Before,
the context rounded such results to an exponent other than -places:
Decimal.round(Decimal.new("1e40"), 2) returned
1.000000000000000000000000000000000E+40, and Decimal.round("1.5", 7000)
returned 1.500000000000000000000000000000000. ±Infinity was returned
unchanged.
Make Decimal.compare/3 exact. It rounded num1 ± threshold to the
context precision, so numbers near the threshold compared wrong:
Decimal.compare(1, "1.000000000000000000000000000000002", "1.5e-33")
returned :eq although they differ by 2e-33. It no longer sets
:inexact or :rounded, accepts -0 as a threshold, and raises
Decimal.Error instead of CondClauseError for a NaN when
:invalid_operation isn't trapped.
Make Decimal.compare/2 raise Decimal.Error for a NaN operand even when
:invalid_operation isn't trapped, instead of returning the NaN. Comparing
a NaN with ±Infinity now raises too, instead of returning :lt or :gt.
Make Decimal.Context.set/1, Decimal.Context.with/2 and
Decimal.Context.update/1 raise ArgumentError for an invalid context: a
precision that isn't a positive integer, an unknown rounding mode, an
emax or emin that is neither an integer nor :infinity, or an emin
above emax. A precision of 0 silently gave wrong results
(Decimal.add(1, 1) returned 0E+1), and the other cases failed with
FunctionClauseError in the next operation.
Make Decimal.to_integer/1 raise ArgumentError for NaN and ±Infinity,
like Decimal.to_float/1, instead of FunctionClauseError.

Configuration

📅 Schedule: (UTC)

Branch creation
- At any time (no schedule defined)
Automerge
- At any time (no schedule defined)

🚦 Automerge: Enabled.

♻ Rebasing: Whenever PR is behind base branch, or you tick the rebase/retry checkbox.

🔕 Ignore: Close this PR and you won't be reminded about this update again.

If you want to rebase/retry this PR, check this box

This PR has been generated by Mend Renovate.

This PR contains the following updates: | Package | Type | Update | Change | |---|---|---|---| | [decimal](https://hex.pm/packages/decimal) ([source](https://github.com/ericmj/decimal)) | prod | minor | `~> 3.1.0` → `~> 3.2.0` | --- ### Release Notes <details> <summary>ericmj/decimal (decimal)</summary> ### [`v3.2.0`](https://github.com/ericmj/decimal/blob/HEAD/CHANGELOG.md#v320-2026-10-10) [Compare Source](https://github.com/ericmj/decimal/compare/v3.1.2...v3.2.0) ##### Enhancements - Make most operations faster. Compared with v3.1.2 on OTP 29, on decimals with up to 20 digits and exponents from -20 to 20, `div/2` is about 8x faster, `mult/2` 6x, `add/2` 5x, `round/3` 4x, `compare/2` 2.7x and `normalize/1` 2x. On money amounts with two decimals, `to_float/1` is about 23x faster, `compare/3` 18x, `div/2` 15x, `rem/2` and `sqrt/1` 4x, `round/3` 2-3x, `add/2`, `sub/2`, `mult/2`, `div_int/2` and `compare/2` 2-2.7x, and parsing 1.4x. `from_float/1` and `cast/1` of a float are about 3.4x faster and `cast/1` of an integer 3x. `to_string/1` is unchanged. - Add the `:subnormal` and `:clamped` signals from the General Decimal Arithmetic spec. `:subnormal` is signalled for every result whose adjusted exponent is below `emin`, even an exact one, and `:underflow` only when such a result is also inexact. A zero result's exponent is kept between `emin - precision + 1` and `emax`, the range of nonzero results, and `:clamped` is signalled when it has to move: `Decimal.mult("0e-3500", "0e-3500")` returns `0E-6176` instead of `0E-7000`. This also applies to the zero results of division by infinity, of `div_int/2` and `div_rem/2` with a smaller dividend, and of `sqrt/1`. A subnormal result that rounds to zero signals `:clamped` too. ##### Bug fixes - Fix `Decimal.div/2` rounding some inexact results the wrong way. Under `:half_even` and `:half_down`, a result just above a tie was rounded as a tie, and under `:ceiling`, `:floor` and `:up`, a result whose first discarded digit was 0 was treated as exact even when later digits weren't, so `:ceiling` and `:floor` could return a value on the wrong side of the true quotient. About 5% of random divisions at the default precision were affected. `:inexact` was also missing when the remainder was a power of ten. - Fix `:up` rounding changing exact values. A result whose discarded digits were all zero was still rounded away from zero: with `precision: 1` and `rounding: :up`, `Decimal.add(0, Decimal.new("9.0"))` returned `10` instead of `9`, and `Decimal.round(Decimal.new(0), -1, :up)` returned `1E+1` instead of `0E+1`. - Fix rounding carries leaving one digit more than the precision. Rounding `9.99` to precision 2 returned `10.0` instead of `10`, and `Decimal.div(95, 10)` at precision 1 returned `10` instead of `1E+1`. This affected every operation that rounds to the context precision, and the results now match the General Decimal Arithmetic spec and Python's `decimal`. - Fix `Decimal.sqrt/1` rounding inexact roots as if the first discarded digit were the whole remainder. At precision 9 under `:ceiling`, `Decimal.sqrt(10)` returned `3.16227766` instead of `3.16227767`, and at precision 2 under `:half_even`, `Decimal.sqrt("1.57")` returned `1.2` instead of `1.3`. The default `:half_up` rounding wasn't affected. Inexact roots now signal `:inexact` as well as `:rounded`. - Fix `Decimal.rem/2` and `Decimal.div_rem/2` computing the remainder from a product rounded to the context precision, which could cancel it out: `Decimal.rem("9999999999999999999999999999999999", "2.000000000000000000000000000000001")` returned `0` instead of `3E-33`. - Fix `Decimal.to_float/1` returning the next float up for odd integers between 2^52 and 2^53, which doubles represent exactly: `Decimal.to_float(Decimal.new(4_503_599_627_370_497))` returned `4503599627370498.0`. - Keep subnormal results instead of flushing them to zero. Since v3.0.0 set the default `emin` to -6143, every result with an adjusted exponent below it became 0 with `:underflow`, even an exact one: `Decimal.div("1e-6140", 10000)` returned `0` instead of `1E-6144`. Such results are now rounded at the exponent `emin - precision + 1` (-6176 by default) with the context's rounding, as IEEE 754 and the General Decimal Arithmetic spec specify, and signal `:subnormal`, plus `:underflow` when that rounding is inexact. A result that rounds to zero keeps that exponent (`0E-6176`). `Decimal.round/3` no longer flushes an input below `emin` to zero before rounding it. - Fix `Decimal.round/3` rounding an input with more digits than the precision twice, first with the context's rounding instead of `mode`: at precision 3, `Decimal.round(Decimal.new("0.4995"), 1, :down)` returned `0.5` instead of `0.4`. - Make `Decimal.round/3` signal `:rounded` when it discards digits of a nonzero coefficient, and `:inexact` when any of them is nonzero, as the quantize operation of the General Decimal Arithmetic spec does. `Decimal.round("1.25", 1)` returned `1.3` with no flags. - Make `Decimal.round/3` signal `:invalid_operation` and return NaN where quantize does: for ±Infinity, when `-places` is outside `emin - precision + 1` to `emax`, and when the result would need more digits than the precision or an adjusted exponent above `emax`. The signal is trapped by default, so these calls now raise `Decimal.Error`. Before, the context rounded such results to an exponent other than `-places`: `Decimal.round(Decimal.new("1e40"), 2)` returned `1.000000000000000000000000000000000E+40`, and `Decimal.round("1.5", 7000)` returned `1.500000000000000000000000000000000`. ±Infinity was returned unchanged. - Make `Decimal.compare/3` exact. It rounded `num1 ± threshold` to the context precision, so numbers near the threshold compared wrong: `Decimal.compare(1, "1.000000000000000000000000000000002", "1.5e-33")` returned `:eq` although they differ by `2e-33`. It no longer sets `:inexact` or `:rounded`, accepts `-0` as a threshold, and raises `Decimal.Error` instead of `CondClauseError` for a NaN when `:invalid_operation` isn't trapped. - Make `Decimal.compare/2` raise `Decimal.Error` for a NaN operand even when `:invalid_operation` isn't trapped, instead of returning the NaN. Comparing a NaN with ±Infinity now raises too, instead of returning `:lt` or `:gt`. - Make `Decimal.Context.set/1`, `Decimal.Context.with/2` and `Decimal.Context.update/1` raise `ArgumentError` for an invalid context: a precision that isn't a positive integer, an unknown rounding mode, an `emax` or `emin` that is neither an integer nor `:infinity`, or an `emin` above `emax`. A precision of 0 silently gave wrong results (`Decimal.add(1, 1)` returned `0E+1`), and the other cases failed with `FunctionClauseError` in the next operation. - Make `Decimal.to_integer/1` raise `ArgumentError` for NaN and ±Infinity, like `Decimal.to_float/1`, instead of `FunctionClauseError`. ### [`v3.1.2`](https://github.com/ericmj/decimal/blob/HEAD/CHANGELOG.md#v320-2026-10-10) [Compare Source](https://github.com/ericmj/decimal/compare/v3.1.1...v3.1.2) ##### Enhancements - Make most operations faster. Compared with v3.1.2 on OTP 29, on decimals with up to 20 digits and exponents from -20 to 20, `div/2` is about 8x faster, `mult/2` 6x, `add/2` 5x, `round/3` 4x, `compare/2` 2.7x and `normalize/1` 2x. On money amounts with two decimals, `to_float/1` is about 23x faster, `compare/3` 18x, `div/2` 15x, `rem/2` and `sqrt/1` 4x, `round/3` 2-3x, `add/2`, `sub/2`, `mult/2`, `div_int/2` and `compare/2` 2-2.7x, and parsing 1.4x. `from_float/1` and `cast/1` of a float are about 3.4x faster and `cast/1` of an integer 3x. `to_string/1` is unchanged. - Add the `:subnormal` and `:clamped` signals from the General Decimal Arithmetic spec. `:subnormal` is signalled for every result whose adjusted exponent is below `emin`, even an exact one, and `:underflow` only when such a result is also inexact. A zero result's exponent is kept between `emin - precision + 1` and `emax`, the range of nonzero results, and `:clamped` is signalled when it has to move: `Decimal.mult("0e-3500", "0e-3500")` returns `0E-6176` instead of `0E-7000`. This also applies to the zero results of division by infinity, of `div_int/2` and `div_rem/2` with a smaller dividend, and of `sqrt/1`. A subnormal result that rounds to zero signals `:clamped` too. ##### Bug fixes - Fix `Decimal.div/2` rounding some inexact results the wrong way. Under `:half_even` and `:half_down`, a result just above a tie was rounded as a tie, and under `:ceiling`, `:floor` and `:up`, a result whose first discarded digit was 0 was treated as exact even when later digits weren't, so `:ceiling` and `:floor` could return a value on the wrong side of the true quotient. About 5% of random divisions at the default precision were affected. `:inexact` was also missing when the remainder was a power of ten. - Fix `:up` rounding changing exact values. A result whose discarded digits were all zero was still rounded away from zero: with `precision: 1` and `rounding: :up`, `Decimal.add(0, Decimal.new("9.0"))` returned `10` instead of `9`, and `Decimal.round(Decimal.new(0), -1, :up)` returned `1E+1` instead of `0E+1`. - Fix rounding carries leaving one digit more than the precision. Rounding `9.99` to precision 2 returned `10.0` instead of `10`, and `Decimal.div(95, 10)` at precision 1 returned `10` instead of `1E+1`. This affected every operation that rounds to the context precision, and the results now match the General Decimal Arithmetic spec and Python's `decimal`. - Fix `Decimal.sqrt/1` rounding inexact roots as if the first discarded digit were the whole remainder. At precision 9 under `:ceiling`, `Decimal.sqrt(10)` returned `3.16227766` instead of `3.16227767`, and at precision 2 under `:half_even`, `Decimal.sqrt("1.57")` returned `1.2` instead of `1.3`. The default `:half_up` rounding wasn't affected. Inexact roots now signal `:inexact` as well as `:rounded`. - Fix `Decimal.rem/2` and `Decimal.div_rem/2` computing the remainder from a product rounded to the context precision, which could cancel it out: `Decimal.rem("9999999999999999999999999999999999", "2.000000000000000000000000000000001")` returned `0` instead of `3E-33`. - Fix `Decimal.to_float/1` returning the next float up for odd integers between 2^52 and 2^53, which doubles represent exactly: `Decimal.to_float(Decimal.new(4_503_599_627_370_497))` returned `4503599627370498.0`. - Keep subnormal results instead of flushing them to zero. Since v3.0.0 set the default `emin` to -6143, every result with an adjusted exponent below it became 0 with `:underflow`, even an exact one: `Decimal.div("1e-6140", 10000)` returned `0` instead of `1E-6144`. Such results are now rounded at the exponent `emin - precision + 1` (-6176 by default) with the context's rounding, as IEEE 754 and the General Decimal Arithmetic spec specify, and signal `:subnormal`, plus `:underflow` when that rounding is inexact. A result that rounds to zero keeps that exponent (`0E-6176`). `Decimal.round/3` no longer flushes an input below `emin` to zero before rounding it. - Fix `Decimal.round/3` rounding an input with more digits than the precision twice, first with the context's rounding instead of `mode`: at precision 3, `Decimal.round(Decimal.new("0.4995"), 1, :down)` returned `0.5` instead of `0.4`. - Make `Decimal.round/3` signal `:rounded` when it discards digits of a nonzero coefficient, and `:inexact` when any of them is nonzero, as the quantize operation of the General Decimal Arithmetic spec does. `Decimal.round("1.25", 1)` returned `1.3` with no flags. - Make `Decimal.round/3` signal `:invalid_operation` and return NaN where quantize does: for ±Infinity, when `-places` is outside `emin - precision + 1` to `emax`, and when the result would need more digits than the precision or an adjusted exponent above `emax`. The signal is trapped by default, so these calls now raise `Decimal.Error`. Before, the context rounded such results to an exponent other than `-places`: `Decimal.round(Decimal.new("1e40"), 2)` returned `1.000000000000000000000000000000000E+40`, and `Decimal.round("1.5", 7000)` returned `1.500000000000000000000000000000000`. ±Infinity was returned unchanged. - Make `Decimal.compare/3` exact. It rounded `num1 ± threshold` to the context precision, so numbers near the threshold compared wrong: `Decimal.compare(1, "1.000000000000000000000000000000002", "1.5e-33")` returned `:eq` although they differ by `2e-33`. It no longer sets `:inexact` or `:rounded`, accepts `-0` as a threshold, and raises `Decimal.Error` instead of `CondClauseError` for a NaN when `:invalid_operation` isn't trapped. - Make `Decimal.compare/2` raise `Decimal.Error` for a NaN operand even when `:invalid_operation` isn't trapped, instead of returning the NaN. Comparing a NaN with ±Infinity now raises too, instead of returning `:lt` or `:gt`. - Make `Decimal.Context.set/1`, `Decimal.Context.with/2` and `Decimal.Context.update/1` raise `ArgumentError` for an invalid context: a precision that isn't a positive integer, an unknown rounding mode, an `emax` or `emin` that is neither an integer nor `:infinity`, or an `emin` above `emax`. A precision of 0 silently gave wrong results (`Decimal.add(1, 1)` returned `0E+1`), and the other cases failed with `FunctionClauseError` in the next operation. - Make `Decimal.to_integer/1` raise `ArgumentError` for NaN and ±Infinity, like `Decimal.to_float/1`, instead of `FunctionClauseError`. </details> --- ### Configuration 📅 **Schedule**: (UTC) - Branch creation - At any time (no schedule defined) - Automerge - At any time (no schedule defined) 🚦 **Automerge**: Enabled. ♻ **Rebasing**: Whenever PR is behind base branch, or you tick the rebase/retry checkbox. 🔕 **Ignore**: Close this PR and you won't be reminded about this update again. --- - [ ] If you want to rebase/retry this PR, check this box --- This PR has been generated by [Mend Renovate](https://github.com/renovatebot/renovate).

rosa added 1 commit

2026-10-17 20:11:34 +00:00

chore(deps): update dependency decimal to ~> 3.2.0

renovate/stability-days Updates have met minimum release age requirement

Details

renovate/artifacts Artifact file update failure

ci/woodpecker/pr/test Pipeline failed

Details

ci/woodpecker/push/test Pipeline failed

Details

f7bf3d817d

rosa scheduled this pull request to auto merge when all checks succeed

2026-10-17 20:11:34 +00:00

rosa commented

2026-10-17 20:11:34 +00:00

Author

Owner

⚠️ Artifact update problem

Renovate failed to update an artifact related to this branch. You probably do not want to merge this PR as-is.

♻ Renovate will retry this branch, including artifacts, only when one of the following happens:

any of the package files in this branch needs updating, or
the branch becomes conflicted, or
you click the rebase/retry checkbox if found above, or
you rename this PR's title to start with "rebase!" to trigger it manually

The artifact failure details are included below:

File name: mix.lock

Command failed: install-tool elixir v1.20.4

### ⚠️ Artifact update problem Renovate failed to update an artifact related to this branch. You probably do not want to merge this PR as-is. ♻ Renovate will retry this branch, including artifacts, only when one of the following happens: - any of the package files in this branch needs updating, or - the branch becomes conflicted, or - you click the rebase/retry checkbox if found above, or - you rename this PR's title to start with "rebase!" to trigger it manually The artifact failure details are included below: ##### File name: mix.lock ``` Command failed: install-tool elixir v1.20.4 ```

rosa merged commit f7bf3d817d into main

2026-10-17 20:11:36 +00:00

rosa deleted branch renovate/decimal-3.x

2026-10-17 20:11:36 +00:00

No reviewers

No labels

No milestone

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

rosa/tragedy!7

No description provided.

Rows
Columns