Skip to content

Latest commit

 

History

History
170 lines (129 loc) · 7.22 KB

CHANGELOG.rst

File metadata and controls

170 lines (129 loc) · 7.22 KB
Raw

Changes from v2.3 to v2.4

With this release, we are no longer supporting Python 2.7 or 3.6. GalSim is supported for the following Python versions: 3.7, 3.8, 3.9, 3.10.

Note

As advertised in the 2.3 release notes, GalSim no longer supports the following legacy options:

  • Python 2
  • TMV for matrices in C++
  • Boost Python for C++ bindings
  • SCons builds

A full list of changes in this release are below. The numbers in parentheses are GalSim issue or pull request numbers where the change was implemented.

cf. https://github.com/GalSim-developers/GalSim/milestone/21?closed=1

API Changes

  • Removed CppEllipse in C++ layer, which had been deprecated since the 1.x series, but we forgot to actually get rid of. (#1129)
  • Removed AstronomicalConstants.h in C++ layer, which we never used. (#1129)
  • Removed AttributeDict, which we had sitting in utilities, but which we have never used. (#1129)
  • Some changes to the C++ Image constructors to include a maxptr value. (#1149)
  • Documented an API change, which was actually introduced in v2.3, that _InterpolatedImage does not recenter the image to (0,0) as InterpolatedImage does. (#1151)
  • Changed SincInterpolant.ixrange to be consistent with the value of xrange, rather than inf. (#1154)
  • Changed galsim.scene namespace name to galsim.galaxy_sample. (#1174)

Config Updates

  • Added Correlated noise type as a generalization of the more specific COSMOS noise type. (#731, #1174)
  • Added galaxy_sample input type with corresponding SampleGalaxy GSObject type. (#795, #1174)
  • Added COSMOSValue and SampleValue value types. (#954, #1174)
  • Allowed template file names to be evaluated using the "$" shorthand notation. (#1138)
  • Added RegisterTemplate to allow the ability to register templates by name. (#1143)
  • Fixed some errors in PhotonDCR usage in the config layer. (#1148)
  • Added option to specify a dtype other than np.float32 for images built by config. (#1160)
  • Fixed inconsistent behavior of image.world_pos in image type=Single. (#1160)
  • Let a flux item for an object with an SED normalize the SED for the bandpass being simulated. (#1160)
  • Fixed some edge cases where the created image could not have the requested wcs. (#1160)
  • Added option to initialize input objects in an InputLoader. (#1162, #1163)
  • Fixed error in returned variance for CCDNoise builder, which was in e- rather than ADU. (#1166, #1167)
  • Changed the way the internal random number sequence works so that running multiple simulations with sequential random seed values doesn't end up with duplicated random values across the two (or more) simulations. (#1169)

New Features

  • Added BaseCorrelatedNoise.from_file class method. (#731, #1174)
  • Added GalaxySample class as generalization of COSMOSCatalog. (#795, #1174)
  • Added methods Image.transpose, Image.flip_ud, Image.flip_lr, Image.rot_cw, Image.rot_ccw, and Image.rot_180. (#1139)
  • Exposed our Si, Ci, sinc, and gammainc functions from C++. (#1146)
  • Added pupil_u and pupil_v to PhotonArray and persist the values thereof when they are computed as part of a phase PSF computation. (#1147)
  • Added Image.depixelize and depixelize=True option for InterpolatedImage. (#1154)
  • Let Bounds.expand take an optional second argument to scale differently in different directions. (#1153, #1155)
  • Added BaseWCS.shearToWorld and BaseWCS.shearToImage along with overloading BaseWCS.toWorld and BaseWCS.toImage to mean the same thing when the argument is a Shear value. (#1158, #1172)
  • Added PupilImageSampler and PupilAnnulusSampler photon operators. (#1176)
  • Added TimeSampler photon operator. (#1178)
  • Added BaseDeviate.as_numpy_generator. (#1067, $1179)
  • Added timeout option to control multiprocessing timeout limit and increased the default. (#1180)

Performance Improvements

  • Change the implementation of the Silicon sensor code to use ~half as many points for the pixels by sharing boundaries between neighboring pixels. (#1118, #1137)
  • Use single precision for Silicon pixel boundaries, which further reduces the memory required for the SiliconSensor implementation. (#1140)
  • Moved some of the logic related to the Silicon sensor to the python layer. This is not per se a performance improvement, but it enables some potential future improvements. (#1141)
  • Let BaseDeviate.generate use multiple threads in C++ layer. (#1177)

Bug Fixes

  • Fixed some cases where HSM would fail to converge for apparently very well-behaved images. (#1132, #1149)
  • Fixed error in InterpolatedImage.withGSParams not correctly updating stepk and maxk if the updated parameters merited it. (#1154)
  • Fixed error in ChromaticSum photon shooting when n_photons is explicitly given. (#1156, #1157)
  • Fixed some rounding errors that could happen when rendering integer-typed images (e.g. ImageI) that could cause values to be off by 1. (#1160)

Changes from v2.4.0 to v2.4.1

  • Fixed the pip installation to include the galsim/share directory, which was missing.

Changes from v2.4.1 to v2.4.2

  • Registered the new photon operators (TimeSampler, PupilImageSampler, and PupilAnnulusSampler) in the config layer.

Changes from v2.4.2 to v2.4.3

  • Fixed error in default nobj calculation for extra_object output when not doing the normal BuildFile processing.

Changes from v2.4.3 to v2.4.4

  • Made it possible to delete items in a config list. (#1183)
  • Fixed error in how input fields check when they are current. (#1184)

Changes from v2.4.4 to v2.4.5

  • Fixed an assert error that would trigger if hsm was run on images with negative stride. (#1185)

Changes from v2.4.5 to v2.4.6

  • Fixed drawImage to work correctly for method=fft when using photon_ops. (#1193)
  • Fixed the proxies used by config Input items to allow access to attributes. (#1195)

Changes from v2.4.6 to v2.4.7

  • Fixed the flux scaling of drawReal for objects with non-diagonal jacobian. (#1197, #1198)
  • Added --log_format option to galsim executable. (#1201)
  • Allowed input objects with has_nobj=True to return an approximate number of objects for the initial pass, rather than the eventual exact number. (#1202)
  • Updated the Great3 links to non-broken URLs. (#1205)
  • Added options to InputLoader to make inputs with AtmosphericScreens work properly. (#1206)
  • Only use proxies for input objects if not yet in a multiprocessing context. This had been causing significant slow downs in some imsim runs. (#1206)

Changes from v2.4.7 to v2.4.8

  • Fixed a slow-down in multiprocessing especially when running very many (>10) processes. (#1213)

Changes from v2.4.8 to v2.4.9

  • Fixed a potential segmentation fault that could happen when using photon_ops with FFT drawing. (#1216)

Changes from v2.4.9 to v2.4.10

  • Fixed the Silicon class to handle invalid wavelengths gracefully rather than raise an exception. (#1227)