Skip to content

Windows: adopt the blaspp space-split fix; correct the set -e comment - #199

Draft
mmelnich wants to merge 2 commits into
mainfrom
drop-space-workaround-and-comment-fix
Draft

Windows: adopt the blaspp space-split fix; correct the set -e comment#199
mmelnich wants to merge 2 commits into
mainfrom
drop-space-workaround-and-comment-fix

Conversation

@mmelnich

Copy link
Copy Markdown
Contributor

Summary

Two small items combined, one of them gated on upstream. Draft until icl-utk-edu/blaspp#137 merges; everything else is final.

Closes #188. Closes #195.

Changes

  • Windows: drop the space-in-path staging workaround once blaspp#137 lands #188: the BLAS++ pin advances past blaspp#137 (currently TEMPORARY, pointing at the PR head c3ef942 so CI can prove the result today; flipped to the merge commit when it lands). RandBLAS's provisioner never carried explicit space-free staging — its vcpkg download lands space-free by construction — so the issue's substance here is the pin plus a standing proof: the Windows lanes now deliberately hand BLAS++ the oneMKL import libraries from a directory containing a space (path with spaces), emulating Intel's default install location. The LAPACK++ pin advances past icl-utk-edu/lapackpp#90 for parity (it only matters for --examples).
  • install.sh carries an incorrect comment about set -e and && lists #195: the set -e comment above the examples-command assembly is corrected per the issue's suggested wording — a top-level [[ test ]] && cmd with a false test does not exit; the construct only bites as the last statement of a function.

Ready when

RandLAPACK mirror (its staging removal + Accelerate ILP64): BallisticLA/RandLAPACK#174.

The comment claimed a top-level '[[ test ]] && cmd' with a false test
exits under set -e. It does not: the failing command is exempt as part
of a && list. The construct only bites as the last statement of a
function, where the test's failure becomes the function's return
status. The if-blocks stay; only the explanation changes.

Closes #195.
Advance the BLAS++ pin past icl-utk-edu/blaspp#137 (temporarily its PR
head; final merge commit when it lands) and the LAPACK++ pin past
icl-utk-edu/lapackpp#90 for parity. RandBLAS's provisioner never staged
libraries into a space-free path -- its vcpkg download lands space-free
by construction -- so issue #188's substance here is the pin advance
plus proof: the Windows lanes now deliberately hand BLAS++ the oneMKL
import libraries from a directory containing a space, emulating Intel's
default install location. Closes #188.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

install.sh carries an incorrect comment about set -e and && lists Windows: drop the space-in-path staging workaround once blaspp#137 lands

1 participant