• Fix UBSan error (misaligned pointer from REAL() on zero-length vector) in build_result() when no equality constraints are present (p = 0). Reported by CRAN M1 SAN checks.

  • Add multi-step solver lifecycle API: ECOS_setup(), ECOS_solve(), ECOS_update(), and ECOS_cleanup() allow re-solving problems with changed numerical data (same sparsity structure) without repeating symbolic analysis. Workspace held in an external pointer with GC finalizer; user data vectors are duplicated to prevent corruption from ECOS in-place equilibration.

  • Switch R-level error messaging from stop() to cli::cli_abort() with inline markup ({.arg}, {.cls}, {.code}, {.field}). The cli package is now in Imports.

  • Add lifecycle tests (10 tests covering setup, solve, update, cleanup, and helper functions).

  • Switch ECOS submodule to fork (bnaras/ecos, r-patches branch); remove ecos-2954b2a-changes/ overlay and simplify Makevars.

  • Fix copy-paste bug in A matrix validation (inherits(G,...) was used instead of inherits(A,...)).

  • Enable dimension consistency check (sum(q) + l + 3*e == nrow(G)).

  • Signal an R error instead of returning silent NULL when ECOS_setup or ECOS_BB_setup fails.

  • Fix R_init function name to match package name (R_init_ECOSolveR).

  • Narrow import(methods) to importFrom(methods, as).

  • C solver fixes: wright omega Taylor series bug, NULL dereference and memory leaks in ECOS_BB, integer truncation in pfloat rounding, missing include guards, printf replaced with Rprintf.

  • Cherry-pick upstream NaN bail-out (PR #181) and ECOS_updateData NULL-safety fix (PR #185) from ECOS develop branch.

  • Add unit tests for ecos.control, make_csc_matrix, dimension validation, and plain-matrix A handling. Fix test-bb.R to namespace-qualify Matrix::sparseMatrix.

  • Clean up matrix coercions
  • Fix CRAN issues re: prototypes.
  • Address the issue caused by Matrix 1.3.x: Force coercion to “dgCMatrix” if not so. Make corresponding change in vignette.
  • Fix some awful cut-and-paste error in R/ecos.R
  • Address edge case for ECOS_BB when G is NULL.
  • Fix compilation issue on Ubuntu by modifying Makevars and Makevars.win to use Rscript call to R.home() to figure out include path. (Issue 5)
  • ECOS source fix for header glblopts.h that defined ECOS_NAN using portable R version R_NaN and R_PosInf. This caused the convolution example in CVXR package to fail on 32-bit platforms.

  • Added a convolution test to be specific.

  • ECOS update: Synced underlying library to version 2.0.7.

  • Removed import of Matrix package, added slam interface, contributions from Florian Schwendinger.

  • Tests: Added a number of unit tests based on base C library.

  • ECOS_csolve assumes A and G as NULL if any dimension is 0
  • The dims list is now coerced to integer immediately upon entry to ECOS_csolve
  • ECOS library sources updated to version 2.0.5.
  • Updated Readme
  • Allowed empty vector c, per Anqi’s request
  • Registered .Call entries
  • Changed Anqi’s email
  • Added system requirement of GNU makefile
  • Improved the description of the package
  • Checked more carefully on Linux, MacOS, and Windows
  • First version, thought I had it right for all platforms, but quite wrong.