Release 0.8.4 "Commands" - Optimizations and equation fixups

[c3d]

This release focuses on improving the solver support for the equation library, fixing various bugs found during development of that equation library, and optimizing the garbage collector. It also introduces configurable keyboard layouts.

Features

• ui: The keyboard layout is now configured by a a config/[keymap].48k file that describes which commands are assigned to which key by default. A new setup entry, Load keymap, lets you change the keyboard layout.

• ui: Add four predefined keyboard layouts:

  • db48x.48k is a key layout that is more logical and effective for DB48x, See thread.
  • legacy.48k is the layout used for earlier releases, which swaps the exp and log10 keys relative to the DM42 in 42style.48k.
  • 42style.48k keeps key layout as close as possible to the DM42
  • true42.48k is identical to 42style.48k, but the simulator shows an image of the DM42 keyboard.

• Add various commands such as Edit to directly perform editing operations in a programmatic way.

• units: Implement non-proportional unit conversions, notably temperature conversions like °F to °C. The underlying engine allows arbitrary conversions, including non-linear ones, which would be useful for example for the Dalton temperature scale. However, that capability is not presently used.

• units: Convert temperatures to K in multiplication and division. For example, when computing ⒸR*T, we need the temperature T to be in K even if given in Celsius or Farenheit initially.

• solver: Report underlying evaluation error. For example, if the expression being evaluated reports Inconsistent units, this is what the solver will return instead of No solution?.

• ui: Add configurable interval for busy cursor drawing, BusyIndicatorRefresh. The default is now 50ms, which refreshes the busy cursor more frequently than before, and may be detrimental to battery life and performance. You can restore the previous behaviour by setting a higher value, e.g. 1000ms.

• performance: Automatic cleanup of temporaries to minimize the number of garbage collection occurences. When a complex operation such as exp is performed, there are a number of intermediate results that require storage, and were previously only cleaned up by the garbage collector. They are now automatically cleaned up before the function returns. The same optimization applies to intermediate graphics while rendering equations, notably on the stack. This delivers significant performance improvements for long-running operations: the "floating-point sum test is now about 20% faster on SwissMicros calculators.

• commands: Add GCStats command to show garbage collector statistics.

• Allow CustomMenu to contain the name of a menu, or a program that builds the menu dynamically. An example is shown in the RPL directory of the Demo file.

• Allow CustomMenu to define "vertical" menus, i.e. menus where the items are stacked on top of one another.

Bug fixes

• Fix functions taking real-like unit input. For example, atan(1_cm/1_m) now computes correctly.

• solver: Do not solve system of equations using existing values. The multiple equation solver would incorrectly consider existing values in variables to check if an equation could be used for solving. The heuristic now picks up the equation that requires the smallest number of unknown variables among the available equations.

• Parse x! as factorial of x and not as a x! symbol. The incorrect parsing was due to an ambiguity in the HP48 parser that was resolved in the HP50G. DB48X now behaves like the HP50G and does not allow ! to appear at the end of a name.

• Do not enter the debugger if DebugOnError is set while evaluating an iferr statement. The assumption is that if you try to catch an
  error, you do not intend to debug the code being tested for an error. If this is not the desired behaviour, then an explicit DebugOnError should be inserted in the body of iferr.

• Various unit-related fixes in the equation library.

• Ensure that we don't execute auto-completed catalog commands twice. The recent change that added the auto-completed command to the command-line history also caused the command to be executed on the command-line before being evaluated again from the key.

• Avoid a rare crash when an equation was too big to be rendered graphically and a garbage collection cycle occurred between graphic rendering and text rendering.

• Do not emit error message from Vec→ for vectors containing units. The incorrect error was introduced by the logic detecting polar, cylindrical or spherical vectors.

• Fixed atan2 special cases to always generate the correct angle. Cases where atan2 would generate an exact result would usually result in the wrong angle unit scaling being applied.

• Fixed parsing of tan⁻¹ in expressions.

• units: Skip the =Cycle section for unit definitions. This was causing incorrect unit conversion errors for users who had added common units in the =Cycle section of their config/units.csv file.

• The user-defined units menu no longer list all the built-in units after the user-defined ones.

• Add a missing sqrt in the RelativisticKineticEnergy sample code, and do not compute the kinetic energies for negative values.

• Alias keys is for KeyMap, not Header. Fix typo in identifiers table.

• Avoid a crash in RandomMatrix when hardware floating-point is enabled, due to an incorrect conversion to integer. As a side effect, fixed a couple of minor issues in the conversion to integer values from decimal or binary floating-point values.

• Fix bogus Bad argument type message for V→ when the number of elements was not 2 or 3.

• Fix crash when dividing a matrix by a non-invertible matrix.

• Fix crash displaying non-normal hardware floating-point values.

• Fix crash in vector operations that cause errors, e.g. [1][0] /.

• Parse ubase algebraic expressions, e.g. ubase(1_km), as well as other function-like commands such as size.

• Accept numerical values in ubase and leave them as is.

• Adjust the "next step" computation in the solver to minimize the complexity of conversions and remove unit ambiguity for temperatures.

• Rewrite the Heat Transfer equations to clarify temperature unit used in the computation. Specifically, avoid having a ΔT in a multiplication, where the value would be incorrectly converted to K.

• Fix a problem when a garbage collection while parsing a fraction could cause a large fraction of the subsequent program to be skipped. This normally led to anomalous Inconsistent units messages when this caused a unit such as 254/10_mm for in to be incorrectly parsed as 254/10.

Enhancements

• Major update to the documentation of the equation library, contributed by Jean Wilson.
• equations: Replace °F with °C in equations
• tests: Add support for tests that are known to fail
• tests: Run equation tests with 11 DIG
• tests: Add ability to take screen snapshots on failure
• units: Put temperatures before pressure in Fluids section
• help: Add image for B Field From Finite Wire
• Makefile: When using make update, do not keep the temporary .png
• Indicate where to get tac in the build documentation
• Fix README link to browser version
• Disable DebugOnError by default, since it confused new users
• Enhance the test suite so that it looks up the keys to use.
  This makes the test suite more readable, e.g. we have ID_exp in
  the test instead of D (the D key being where exp was on the
  original layout), and it makes the test suite layout-independent,
  paving the way for reuse on other hardware (e.g. HP50G).

---

This discussion was created from the release Release 0.8.4 "Commands" - Optimizations and equation fixups.