LAST production releases - EranOfek/AstroPack GitHub Wiki

This page lists the AstroPack release tags that run in production on the LAST computers. For each tag it gives a short description and the changes since the previous tag (the "delta").

How the tags are used. A release is an annotated tag on main, named lv<major>.<minor>.<patch>. The LAST computers do not follow main: they are fixed at one tag. The tag is set in the LAST_tools file files/github-repos.gledmagicwater (branch:main,tag:<tag>), so last-fetch-from-github keeps them at the tag. A change reaches production only when a new tag is made and that file is updated. See issue #1266.

Tag Commit Date In production Main changes
lv1.0.3 71cbf5380 2026 Oct 14 2026 Oct 14, 16:02 UT, all 20 computers much less memory on very dense fields: Gaia stars for the transient match read per sub-image (#1387)
lv1.0.2 26ba4efe0 2026 Oct 14 2026 Oct 14, 10:26 UT, all 20 computers correct barycentric JD and velocity in the transients database (#1395)
lv1.0.1 4ec180318 2026 Oct 13 2026 Oct 13, 19:35 UT, all 20 computers transients written to the database again (#1396); bright-sky images no longer rejected by the histogram check (#1179)
lv1.0.0 456c80f68 2026 Oct 11 2026 Oct 11, 18:10 UT (18 computers), last03w/last04w on Oct 13 first release

lv1.0.3

Commit 71cbf5380 on main, 2026 Oct 14, 4 commits after lv1.0.2 (plus 2 merges). Release note: #1266.

Changes that affect production processing

  • pipeline.last.pipes.pipelineII reads the Gaia stars for matching transient candidates for each sub-image (a cone of the sub-image half-diagonal + 300") instead of one catalog for the whole visit. Near the Galactic Centre the visit-wide query held 8.4 million stars and briefly needed about 20 GB more memory. Tested on 7 visits with lv1.0.2 and the new code: all products identical; peak memory 87.7 -> 63.5 GB (field 524) and 80.2 -> 63.5 GB (field 608); a normal field is not slower (#1387). (Ruslan, 0d7a36375)

Other changes (not used by the production pipeline)

  • imProc.stack.coadd_Proper: missing pixels are no longer counted as real measurements (#1399). (Dana, 3d85f884a); imUtil.properCoadd.properCoaddLinear updated. (Eran, 5398854eb) Production coadds use coadd_WRobust.
  • pipeline.last.reference.buildRefImages and imProc.stack.stitchCrops: reference building recovers the node number of old coadds and finds coadds named with a different time. (d5268dd84)

Deployment. The computers were moved to the tag at 16:01 UT on 2026 Oct 14 and each pipeline was restarted when idle (39 of 40 by 16:04 UT; 01.06.03 at the end of the visit it was processing). LAST_tools names lv1.0.3 since 607e04a (last-tool 1.0.9783.28891). The newer commit 399e48b20 (#1377) was not tested and is not in this release.

lv1.0.2

Commit 26ba4efe0 on main, 2026 Oct 14, 3 commits after lv1.0.1.

Changes that affect production processing

  • celestial.time.barycentricJD gave wrong results whenever it was called with more than one time or position: the light-travel-time correction and the barycentric velocity were divided by about the square root of the number of rows. imProc.db.insertCatalog calls it for a whole catalog at once, so the BJD and BARYVEL columns written to the database were wrong (in last.diff_src about 5 times too small). Single-value calls were always correct. Rows already in the database are not corrected by this release (#1395). (26ba4efe0)

Other changes (not used by the production pipeline)

  • pipeline.generic.procCoadd: a proper coadd is now always treated as background-subtracted by the source extractor (#1394). Production coadds use coadd_WRobust. (Dana, 3a79cd2bc)
  • imProc.calib.fitPhotCalibTrans: bands of other missions can be calibrated through a Mission setting (#1393). The default mission is LAST, and the LAST results do not change. (Dana, 394bcdf90)

Deployment. Deployed on 2026 Oct 14 with the pipelines stopped and restarted: last01e at 10:21 UT, the other 19 computers at 10:26 UT. LAST_tools names lv1.0.2 since 6338aa1 (last-tool 1.0.9783.18831).

lv1.0.1

Commit 4ec180318 on main, 2026 Oct 13, 24 commits after lv1.0.0 (plus 5 merges). Release note: #1266.

Changes that affect production processing

  • Transient candidates are written to the database again (#1396). Nothing had been written since 2026 Sep 18. Three code problems were fixed:
    • transient writing was not turned on in production;
    • the native ClickHouse connector failed on every insert (empty schema, numbers in the wrong column type);
    • NaN values made the insert fail. Re-reductions of visits older than 2 days do not write. (bbed03abf, 0670d4633)
  • The histogram check (imUtil.image.histAnomaly) is skipped when the sky is above 3000 ADU. On a bright (moonlit) sky it gave false alarms and rejected good images (#1179). (4ec180318)
  • imProc.sub.smearTemplate (used by pipelineII) uses the derived smear in preference to the measured one, and its S/N column is fixed. (Ruslan, 6da0e0ad2)

Other changes (not used by the production pipeline)

  • imProc.stack.coadd_Proper: the background is measured on the normalized coadd scale (#1329), and the PSF star count is set so the seeing is measured (#1390). Production coadds use coadd_WRobust. (Dana)
  • New ts.fitps functions (power-law and damped-random-walk power-spectrum fits) and ts.rand.randomPowerLawLightCurve. (Eran)
  • Database insert speed tests: imProc.db.insertCatalog gets a SkipDbInsert option (default false), plus a separate insertCatalog_perf, performance-logging functions, scripts and notes under pipeline.last.insertDB. (Chen)

Deployment. 0670d4633 (the #1396 fix) was deployed on 2026 Oct 13, 18:55 UT. 4ec180318 followed at 19:29–19:35 UT, and the tag was then made on that same commit. LAST_tools names lv1.0.1 since 6f46e44 (last-tool 1.0.9782.35472).

lv1.0.0

Commit 456c80f68 on main, 2026 Oct 11. This is the first release (#1266): the code that dev1 had when it was merged into main on 2026 Sep 26, with later fixes up to Oct 11. It runs the v1 pipeline (PipelineDemon) on all cameras. Its last changes before the tag:

  • the memory and file-handle leaks in the FITS writer, the FITS readers and sliceGaussianProfile are closed (#1388);
  • pipelineI frees the full-frame images after the cut into sub-images, about 9 GB less memory (#1367);
  • pipelineI runs maskHoles, fwhm and fitPhotCalibTrans on the parallel workers, about 25 s faster per visit (#1256).

There is no earlier tag to compare it with. For the changes made before Oct 11 see the New functions and changes log.

⚠️ **GitHub.com Fallback** ⚠️