Automatically choose map data type #62
Comments
|
Map factory creation will likely need to be deferred to a point when source photo parameters are available, i..e. when the source photo FITS file has been opened. This will likely entail moving the map creation code out of the MaRC input file parser to a something like a new Abstract Factory (or other creational design pattern) implementation for MaRC's maps. |
ossama-othman
added a commit
that referenced
this issue
Aug 31, 2018
Type-specific MapCommand instances are no longer created during MaRC input file parsing. Type-specific operations are now deferred until the MapCommand is actually executed. This allows type related decisions to be made internally without user input. Upcoming changes will include the automatic type selection logic. Related to #62.
ossama-othman
added a commit
that referenced
this issue
Feb 1, 2019
* FITS_file: Begin encapsulating FITS file operations. * lib: Corrected auto data scaling comments. Clarified and corrected comments regarding calculation of scale and offset values for MaRC virtual images. The comments incorrectly implied that no loss of significant digits would occur when assigning the scaled and offset floating point values to an integer. Some loss of significant digits will always occur. The goal is to retain as many significant digits as possible without a complete loss of them. * lib: Move scale_and_offset() to SourceImage.h. Make the MaRC::scale_and_offset() function available to other MaRC source images that aren't virtual images by moving it to the SourceImage.h header. * lib: Added sanity checks for cosine images. * Clarify where physical data values are used. * ImageFactory: Validate physical data min/max. * src: Set appropriate min/max for virtual images. Explicitly set narrower minimum and maximum values in MaRC virtual image factories since we know what they are. * src: Re-enable column retrieval in scanner. * Removed need for MARC_HAS_3_PARAM_STD_HYPOT. Since the 3 parameter version of std::hypot() is available in C++17, simply check if the standard C++ __cplusplus preprocessor symbol is defined to a value greater than or equal to 201703L, rather than perform a check at configure-time. This allows us to slightly reduce configure script maintenance and size, as well as to avoid exposing the private <MaRC/config.h> header through the public <MaRC/Mathematics.h> header. * src: Address unused function warning. * Address clang-specific warnings. * src: Improved input file syntax error reporting. * parse: Explicitly mark empty parser rules. Per GNU Bison conventions, mark empty rules using the "%empty" directive to make it obvious that the rule is indeed empty. * parse: Eased requirements for MU and MU0 planes. The MU plane no longer needs to have the sub-solar point specified. It was never needed for emission angle calculations. Similarly, the MU0 plane no longer needs to have the sub-observer point and range specified. They were never needed for incidence angle computations. No changes to the PHASE plane were made. Fixes #59. * lib: Add MaRC::optional<> implementation. The new MaRC::optional<> class template implements a subset of the standard C++17 std::optional<>. It lacks conversion constructors, the emplace() method, the std::hash() specialization, etc. It exists to allow MaRC to use std::optional<>-like functionality without forcing it to require C++17 (std::optional<> was introduced in C++17). However, if C++17 is enabled through compiler flags, the compiler supplied std::optional<> will be used as the underlying implementation. MaRC::optional<> is only meant for use until MaRC's transition to C++17 occurs. As such, MaRC::optional<> is considered deprecated. Do not count on it to exist once MaRC requires C++17 (not expected any time soon). * lib: Fix problem of single longitude being mapped. Corrected a problem where MaRC's Simple Cylindrical projection could not handle the case where lower and upper longitudes in the map were the same, signifying a 360 degree range. For example, passing 0 as the lo_lon and hi_lon longitude values in the MaRC::SimpleCylindrical constructor results in a map with a single longitude 0 being mapped across all map pixels. Fixes #95. * src: Progress console implementation is now final. MaRC::Progress::Console is not inherited any other class. Declare it as "final" to allow the compiler to perform additional optimizations. * src: Silence unused argument warning. * configure: Bump the version to 0.10a (alpha). * Initialize map to provided blank value. Initialize integer typed maps to the provided blank value, i.e. DATA_BLANK in the MaRC input file, rather than zero. This allows FITS image readers to easily identify areas of integer typed maps that lack data. Fixes #94. * Map_traits: No need to add 1 to min value. * Map_traits: Simplify mininum()/maximum() traits. Leverage std::min() and std::max() and a comment to make it obvious what the code is doing. Removed unnecessary 'float' specialization. * lib: Determine map minimum and maximum only once. * Map_traits: Minor Doxygen documentation updates. * Moved to_unit_vector() and magnitude() to Vector. It makes more sense for the MaRC::to_unit_vector() and MaRC::magnitude() functions to be members of the MaRC::Vector<> class template. Changed accordingly. * lib: Move scale_and_offset_impl() to details dir. Move the MaRC::scale_and_offset<>() implementation details, i.e. the related functors and specializations to the new `details' directory to make it clear that they are not part of the public API. * lib: Minor fixes to lat-lon-at-center code. Fixed some issues with the viewing geometry code that compute rotation matrices when the latitude and longitude at the center of the image are given. Vectors in different coordinate systems were be compared. Relates to #29. * lib: Make BodyData constructor constexpr. * lib: Minor Doxygen documentation improvements. * lib: Remove useless assignment operators. Remove the assignment operators that accept an initializer list from the MaRC::Vector and MaRC::Matrix class templates. They are never used since C++ copy-list-initialization (e.g. Vector foo = {1, 2, 3}) is performed through the constructor that accepts an initializer list. * Vector: Add parameter pack constructor. Added a constexpr constructor to the MaRC::Vector<> class template that accepts parameter pack. This allows for direct initialization of a Vector like so: MaRC::Vector<double, 3> foo(1, 2, 3); and: constexpr MaRC::Vector<double, 3> foo(1, 2, 3); The direct-list-initialization constructor, e.g. foo{1, 2, 3}, still exists but it is not constexpr. * ViewingGeometry: Directly initialize Vectors. Simplify MaRC::Vector initialization by passing element values directly to the constructor. * Vector: Simplify ranged-based for loop in impl. There was no need to use "*this" as the range expression in the range-based for loop in the MaRC::Vector operator*=() and to_unit_vector() implementation. Just use the underlying fixed-size array directly. Added a note about why we don't do the same for the MaRC::Matrix::operator*=(). * Vector_Test: Fix constexpr related build error. Some older compilers don't allow a list-initialized std::initializer_list<> to be declared constexpr since the braced-initializer-list wasn't considered constexpr. * tests: Add Vector and Matrix element access tests. * MapFactory: Corrected blank value range check. Verify that the user-supplied blank value, not the MaRC default blank value, fits within the map data type. * lib: Clarify need for floating point type in comp. Clarify that the MaRC::almost_equal() and MaRC::almost_zero() function templates require a floating point type by making the enabling condition rely on std::is_floating_point<> rather than !std::numeric_limits<>::is_integer. * Removed redundant braces from initializer lists. * Match include directives against install path. Headers are installed in $prefix/include/marc not $prefix/include/MaRC. Adjust the include directives in the source accordingly. Fixes #97. Added an installcheck-local rule that runs the installed `marc' binary against the main MaRC test input file. This change really belonged in separate commit, but we'll let it slide this time. * tests: Implemented tests for mu, mu0 and cos(phi). Implemented oblate spheroid tests for the cosines of the emission, incidence and and phase angles that calculate that confirm that results of the MaRC::OblateSpheroid::{mu, mu0, cos_phase}() methods return correct results, respectively. * FITS_file: Support CFITSIO <= 3.39 Drop use of the fits_get_key_strlen() and fits_read_string_key() functions that were introduced in CFITSIO 3.39 to support older CFITSIO installations. * src: Increase precision of stringified double. * src: Removed MapCommand_T class template. Type-specific MapCommand instances are no longer created during MaRC input file parsing. Type-specific operations are now deferred until the MapCommand is actually executed. This allows type related decisions to be made internally without user input. Upcoming changes will include the automatic type selection logic. Related to #62. * lib: Add include guards in template source. * src: Work around old non-constexpr std::min(). Some older implementations of the std::min() function are not constexpr despite the fact C++14 support is explicitly enabled. Rely on the ternary operator instead. * Leverage C++14 attribute lists where applicable. * lib: Remove deprecated MaRC::library_version(). * MapCommand: Make protected members private. The MaRC::MapCommand class is no longer a base class, meaning there is no need to make it some of its members protected to allow access by subclasses. Just mark those members as private. * Support older Autoconf code coverage macros. Work around the break in backward compatibility of the AX_CODE_COVERAGE Autoconf Archive macro. Both the old and new versions of the macro are now supported. * doc: Minor documentation clarification. * Fixed code coverage build regression. * README: Replace 'tarball' with 'tar archive'. * README: Corrected header file installation path. * m4: Reintroduced ax_code_coverage.m4 file. Some very old autoconf-archive installation lack the ax_code_coverage.m4 file. Reintroduce the latest version. * lib: Added note about grid interval validation. * Mercator: Refactored redundant calculation. * SimpleCylindrical: Make lat bounds const. Initialize the Simple Cylindrical lower and upper map latitudes in a way that allows their corresponding class members to be const. This should improve the potential for additional compiler based optimizations. * Revert "Mercator: Refactored redundant calculation." This reverts commit 691b67565343af32235cd3c1359c95383c3df0bf. Accuracy seems to drop slightly, causing the Mercator test to fail. Revert for now. * Replace use of spdlog with fmt library. Source files that included <spdlog/spdlog.h> resulted in non-trival increases in object file size. Drop spdlog for now, and use the fmt library (used by spdlog under the hood, too) to satisfy MaRC's simple logging requirements, for now. Fixes #96. * MapFactory_t: Do not rely on NDEBUG definition. There is no guarantee that the NDEBUG preprocessor symbol will be defined in the MapFactory_t.cpp template source file since the private <marc/config.h> header is not, and should not be, included. Doing so would expose that header to the user. Removed code that relied on that preprocessor symbol definition. * MapFactory_t: Corrected blank value range check. The check that attempted to verify that the user supplied map blank value fit within the map data type range always succeeded since the user supplied value was explicitly cast to the map data type before the check. Cast to the map data type after the check to allow the test to fail when it should. * tests: Improve MaRC test harness. Leverage Automake's parallel test log compiler feature to simplify how MaRC input files are tested. MaRC input files now need only be listed in the "TESTS" Makefile variable instead of creating a custom wrapper script that runs MaRC on each input file. * parse_scan: Simplify confusing error message. * tests: Add negative tests. Expanded the test suite by adding tests that are expected to fail due to bad values in MaRC input files. This exercises the error handling in the MaRC input file parser. * parse: Input file parser improvements. Make "GRID" optional in input files (defaults to "NO"), and only allow specification of the grid interval if the grid was enabled. Added missing validation of ARCSEC_PER_PIX values. Minor error message tweaks. * Distribute 'opt' dir to fix distcheck. * m4: Update autoconf archive macros. * lib: Updated several constants. Updated several physical constants and conversion factors to their latest values, as revised by the BIPM and IAU. * Improved static constants documentation. * src: Clean up MaRC::FITS::traits class template. Only specializations of the MaRC::FITS::traits<> class template are ever expected to be used since there is no general way to define traits for all CFITSIO image and type codes. Only forward declare the generalized class template to prevent potential errors related to unhandled types from occurring in the future. Drop the supports_blank_keyword trait in favor of a std::is_integral<> based check. There was no real need for that trait since it easy to determine if the type in question is an integer at compile-time, and thus support blank FITS values. Corrected multiple Doxygen warnings and issues. * lib: Added notes about physical constant updates. * Fixed all Doxygen warnings. * More Doxygen documentation corrections. Prevent automatic linking to the MaRC and FITS namespaces where appropriate, e.g. when referring to the MaRC program or library, and the FITS standard. Replaced incorrect usage of the Doxygen @internal special command with @note. * src: Begin consolidating map parameters. Moved several map parameters found in the parser and the MaRC::MapCommand class to a separate MapParameters class. This allows a single object to maintain map parameters instead of having duplicates of those parameters in multiple locations. Reduced number of global variables in the parser. * lib: Move scale_and_offset() to its own header. <marc/SourceImage.h> wasn't really the best place to put the MaRC::scale_and_offset() function template. Just place it in its own header. * Various doc corrections and improvements. Corrected class and struct Doxygen documentation in the MaRC library so that the header path includes the MaRC package include directory, e.g. <marc/BodyData.h> instead of <BodyData.h>. Minor Doxygen documentation corrections and improvements, including file documentation that got mixed into macro documentation, and adding template parameter documentation where appropriate. * tests: Removed marc_program_test.sh With the recent the changes to tests/Makefile.am that allowed MaRC input files to be listed directly as tests there is no longer any need for the marc_program_test.sh shell script. Remove it. * lib: Generalize image inversion functions. Generalize the MaRC::invert_samples() and MaRC::invert_lines() functions so that they work on any STL style container with the appropriate iterators, as well as basic arrays. * lib: Clarify map data values as "physical". * src: Improved file parse error reporting. Do not ignore syntax errors in the user configuration file. Corrected typo that treated scanner initialization failure as a success. * Read and set the unit in PhotoImages. If available, read the unit name from the photo FITS file, and store it in the PhotoImage for potential use in the map FITS file. * lib: Minor documentation tweak. * Leverage C++11 override where appropriate. Use the C++11 "override" specifier on virtual functions that should override a base class virtual function to help prevent inheritance related errors (e.g. function overload vs override). * src: Improve number parse error reporting. Report underflow and overflow during conversion of stringified numbers to a numerical value to the user. Make division-by-zero error detection in the parser fatal. * tests: Expand DATA_BLANK tests. * Get fixed width integer types "std" namespace. Since we're getting the fixed width integer types from the standard C++ library <cstdint> header explicitly use the "std" namespace. * Consolidate "blank_type" type alias declarations. Use the "MaRC::plot_info::blank_type" alias in the MaRC library as the underlying blank type used in the MaRC program. This prevents potential mismatches. Replace the underlying blank integer type from int64_t to intmax_t in case the platform supports an integer with greater than 64 bits. The blank integer value will still be cast as needed depending on the map integer data type. * lib: Fix ortho projection longitude validation. * lib: Simplify Orthographic constructor. * tests: Expand input file test. * src: Fix BLANK value access. The BLANK value should have the same type as the map projection. Furthermore, use the address of the extracted blank value, not the address of the MaRC::optional<> containing the blank value. * tests: Minor Mercator test doc corrections. * lib: Add constexpr OrthographicCenter constructor. Added a constexpr MaRC::OrthographicCenter constructor to allow constexpr instances to be declared. * tests: Added new Orthographic projection test. * Fix artifacts in the orthographic projection (#102) * lib: Fixed MaRC::quadratic_roots() regression. Fixed a regression in quadratic roots calculations that occurred when the "b" coefficient of a quadratic equation being solved was zero. This addresses problems that manifested in the Orthographic projection, coordinate transformation calculations, etc. * lib: Improve Orthographic projection precision. Addressed a potential catastrophic cancellation in the Orthographic projection in a calculation of the form (x*x - y*y). Use the equivalent form (x-y)(x+y) where cancellation due to subtraction is benign. * plot_info: Make construction more convenient. MaRC::plot_info now has multiple constructors to handle three blank value cases: (1) integer typed blank values, (2) floating point blank values (these values are always ignored), and (3) no blank value provided. They were added to simplify construction of MaRC::plot_info in generic code. * tests: Orthographic projection test now passes. The orthographic projection test was failing due to use of a map data type that could not provide the precision necessary to support the floating point equality strictness (small ulp value) in the test. Force the use of either a floating point map data type, or an integer type that is at least 64 bits wide. * README.md: Update Travis CI URL. * lib: Drop zero check in MaRC::quadratic_roots(). Drop the newly added "b == 0" check in MaRC::quadratic_roots() in favor of a branchless implementation that relies on a newly added MaRC::sgn() function. The new MaRC::sgn() function returns the sign of a real number. It differs from the signum() function in that it interprets the value zero as having a positive sign, i.e. 1, instead of 0. This behavior is required for the numerically stable form of the quadratic formula implemented in MaRC::quadratic_roots(). * OblateSpheroid: Minor code simplification. Simplify implementation of MaRC::OblateSpheroid::M() and MaRC::OblateSpheroid::N() so that they are easier to read. * src: Moved MaRC::strerror() to its own header. * src: Move input file number conversion to parser. Handle MaRC input file related string to number conversion in the parser to make it possible to convert 64 bit integers in string form directly to the underlying integer type instead of using variable of type double as an intermediary. Addresses an issue where the smallest and largest 64 bit integers could be not be passed from the scanner to the parser since the largest integer that can be stored in a 64 bit IEEE 754 floating point value (type double) without loss of precision is +/- 2^53. * lib: Add missing include guards to <marc/Log.h>. * plot_info: Disallow supplied NaN minimum/maximum. * tests: Exercise 64 bit blank value. * src: Defer setting source image extrema. Postpone setting source image extrema until the image is created. This addresses an issue that prevented 64 bit integer data from being mapped. * src: Fixed blank type mismatch when verifying. Rather than type double, verify a blank value that has the same integer type used by MaRC. This allows 64 bit integer values to be properly verified and formatted on error. * parse: Renamed strtonum() to from_string(). * src: Mark BSCALE and BZERO as unset by default. Do not set BSCALE and BZERO map parameters by default since we only want to add them to the map FITS file if they were explicitly set. * Minor doxygen documentation fixes. * Matrix: Corrected typo documentation note.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
MaRC should automatically choose the map data type based on the data type in the source photo FITS file if a type is not explicitly chosen by the user. For example, if a source photo is comprised of 16 bit signed integers (FITS "short"), the generated map should also contain 16 bit signed integers.
The text was updated successfully, but these errors were encountered: