MATLAB Update towards R2020b & Runtime

[ocehugo]

We already reached a 2-year difference with recent/upstream MATLAB versions. Maybe time for an update!? I compiled a curated list of changes that matter for us below. Not listed are several bugfixes, related changes in toolboxes, and some updates to NetCDF library. There were some big changes in 2020, but important ones also happened in 2019.

The most exciting changes are in the R2019 version which would help immediately. The deeper introspection and function argument validation allowed in 2019b is particularly useful. R2020 also got some new things that may help in the long-run or in new features (or even help on the modernization of the UI from GUIDE to app-designer).

Did anyone had a go with 2020b? Any voices pro/against an update over the next quarters? Also, we got a case of graphical issues with 2019 within the UI during the workshop day. Please link here anything related to versions beyond 2018b so we can foresee what we will need to do.

R2019a:

• readmatrix,readvars,readcell
  • improve csv/xls reads, parsers
  • may deprecate/reduce code size in some parsers.
• append function
  • named append of strings, instead of [astr anotherstr]

R2019b:

• function argument validation body
  • explicit declaration of argument types - much, much better function declaration
  • function body now can be declared with arguments, a-la classes properties.
• allow deeper introspection of functions
  • Would allow abc().mymethod, which is very handy.
• matches function
  • better fit than strcmp and would help in corner case handling.
• size function allows the respective index sizes to be extracted
• hexadecimal support (maybe good for binary parsers).

R2020a:

• bin2dec / hex2dec
  • cleaner binary parsers.
• direct access to s3 (and AODN files!)
  • get auxiliary files directly (direct access to long-timeseries products, direct access to previous deployments)
  • may help us to not distribute some files anymore (e.g. templates, DB files !?)
  • Get auxiliary data easily, no need for redistribution (e.g. get/update climatology file).
  • may help in auxiliary plots, comparisons, graphics, climatology, etc.
• if/while/switch evaluations with cast conversion
  • cleaner code, no inline conversions necessary.
• improved graphics functions and utilities
  • ChartContainer
  • improved xlabel and friends
  • better axis interactions
  • data tip marker positions)
• improved UI drawing performance (e.g. buttons)
• faster transpose, interp1, assert functions
• detrend ignore nans (may help in PP/QC).
• readTable automatic detection on import (may improve/deprecate CSV parsers)
• FlameGraph performance analysis.

R2020b:

• Cast Conversion at Assignment
  • C(k) = A (transforms A into C type implicitly at assignment).
• dataStore
  • Allow some declarative access patterns instead of implicit indexing/cutting/dicing.
  • may transform the way we access some axis/dimensions on PP/QC/display data.
  • May help in large datasets.
• Better methods in handling strings
  • improvements text reading and parsers mostly.
  • pattern func (similar to regexp but flexible)
  • extract func (direct extraction of substrings)
• More built-in type-validation methods
  • mustBeFloat, mustBeVector, mustBeInRange, mustBeText,etc. (deprecate some existing).
• scatter transparency, padded axis functions.

[sspagnol]

We (AIMS) don't have an issue with updating to R2020b. And you can also add matlab upgraded netcdf support to v4.7.3. Now if only they had something like xarray.

[anaredd]

At SARDI we don't have any issues with upgrading to R2020b either. It looks useful

[ocehugo]

We (AIMS) don't have an issue with updating to R2020b. And you can also add matlab upgraded netcdf support to v4.7.3. Now if only they had something like xarray.

@sspagnol , It's not xarray, but with nc_flat you can access global attributes and variable attributes by name and with autocompletion :). Better than nested for loops with string (id) checks (picking).

At the same time I did nc_flat I tried a class to lazy-load variable content ala xarray, but matlab classes get/set methods inner workings required more hacking to work efficiently. The functional introspection abc().mymethod could be useful to materialise this and join with nc_flat into a more generic class.