[vaibhavsrv]

Why this matters

In ods/installers/lib/amd-topo.sh, amd_memory_type() classifies whether an AMD GPU topology exhibits discrete or unified memory by performing integer division in Bash arithmetic ($(( gtt_bytes / 1073741824 )) and $(( vram_bytes / 1073741824 ))). When topology discovery encounters missing sysfs entries, unpopulated driver nodes, or placeholder text such as "N/A", Bash under set -euo pipefail halts with unbound variable or arithmetic syntax errors, terminating multi-GPU topology discovery before fallbacks can be evaluated.

This patch adds surgical regex guards [[ "$vram_bytes" =~ ^[0-9]+$ ]] || vram_bytes=0 and [[ "$gtt_bytes" =~ ^[0-9]+$ ]] || gtt_bytes=0 in amd_memory_type(). Any non-numeric or missing input safely defaults to 0 bytes and cleanly falls back to "discrete". Existing APU threshold logic and valid topology detection remain untouched.

Validation

• Baseline: Invoking amd_memory_type "invalid" "N/A" failed with line 50: N: unbound variable and non-zero exit code.
• Post-fix: Non-numeric and empty inputs cleanly return discrete with exit code 0; valid discrete and unified byte counts continue to evaluate correctly.
• AMD topology detection suites: 4 passed. New-test Ruff, ShellCheck, and diff checks pass; new regressions wired into Linux CI.

Overlap check

• fix(pixel-inference): validate top_logprobs integer bounds #6322 (fix(pixel-inference): validate top_logprobs integer bounds) — inference API parameter bounds; zero shell installer overlap.
• fix: export AMD GPU indices safely with defaults #6302 (fix: export AMD GPU indices safely with defaults) — GPU index export environment formatting, separate from memory type topology classification.
• fix(amd-topo): format vram with LC_ALL=C for decimal-comma locales #6076 (fix(amd-topo): format vram with LC_ALL=C for decimal-comma locales) — decimal formatting in topology JSON output, adjacent helper.

Risk / AI disclosure

AI-assisted investigation, implementation and CLI regressions. This strengthens a shell arithmetic input guard, not runtime admission. Independent human review and platform/runtime qualification remain gates. No running configuration, deployment or upstream merge changed.

Follow-up integration evidence

Composed with #6076 at 5662f0d without conflicts. Production and test diffs passed together; AMD topology classification checks remain intact.
Backlog composition was local-only (production/test diffs, excluding workflow/Makefile wiring); it is not an upstream merge or independent human approval. Declared live-review gates remain open.

[Lightheartdevs]

Thanks for this contribution. public-beta was promoted into main on 2026-09-24 and no longer receives changes, so we're closing pull requests that target it. This isn't a judgment on the change itself. If it's still needed, please rebase onto main and open a focused PR. See #7253 for details and the contribution policy.