Milab-6720: fix tsv separator, fix anarci version

.changeset/anarci-with-docker-entrypoint.md

@@ -0,0 +1,23 @@
+---
+'@platforma-open/milaboratories.import-vdj.workflow': patch
+'@platforma-open/milaboratories.import-vdj': patch
+---
+
+Move to software-anarci 1.0.1, which records its docker entrypoint
+
+Region annotation failed on k8s deployments with `sh: 1: ANARCI: not found`
+(exit 127). The conda image puts its environment on PATH solely through
+`ENTRYPOINT ["micromamba", "run", "--prefix", "/conda-env"]`, and a k8s pod spec
+overrides the image entrypoint — so the runner has to re-apply it from the
+software descriptor. Every software-anarci up to 1.0.0 records
+`docker.entrypoint: []`, because the package-builder that published them did not
+read the built image's entrypoint back. Nothing re-applied the wrapper, ANARCI
+was never on PATH, and the step died before it started.
+
+1.0.1 was rebuilt with a package-builder that reads `.Config.Entrypoint` from the
+image, so its descriptor carries the micromamba wrapper and the k8s runner
+reconstructs the right command. `^0.0.3` is exact for a 0.0.x range, so the pin
+could never pick the fix up on its own.
+
+No ANARCI behaviour changes between these versions — 1.0.0 was a plain release
+and 1.0.1 was "update build deps".

.changeset/separator-name-not-character.md

@@ -0,0 +1,21 @@
+---
+'@platforma-open/milaboratories.import-vdj.workflow': patch
+'@platforma-open/milaboratories.import-vdj.column-profile': patch
+'@platforma-open/milaboratories.import-vdj': patch
+---
+
+Pass the column-profile separator by name, not as a tab character
+
+Loading a TSV failed on server deployments with `TypeError: "delimiter" must be a 1-character
+string`, while the same file loaded on a desktop backend. The block was passing a real tab as an
+argv element. Desktop runners exec argv directly, so the tab arrived intact; the k8s and
+google-batch runners serialise the command with Go's `%q` and re-run it through `sh -c`, where
+the tab has already become the two characters `\` and `t` and stays that way. `csv.reader`
+rejects a two-character delimiter.
+
+Prerun now sends `tab` or `comma` and `main.py` maps the name back to the character, so only
+plain words cross the runner boundary. A separator that still arrives malformed now fails with a
+message naming the accepted values rather than a `TypeError`.
+
+The underlying quoting is a backend issue and is unfixed: `toShellCmd` in `util/k8s/template.go`
+uses Go quoting where POSIX shell quoting is needed.

.changeset/software-build-via-block-tools.md

@@ -0,0 +1,23 @@
+---
+'@platforma-open/milaboratories.import-vdj.column-profile': patch
+'@platforma-open/milaboratories.import-vdj.region-annotation': patch
+'@platforma-open/milaboratories.import-vdj.xlsx-to-csv': patch
+'@platforma-open/milaboratories.import-vdj': patch
+---
+
+Build software through block-tools so dev builds produce docker images
+
+The three software packages called `pl-pkg build` directly, which ignores
+`PL_BUILD_CHANNEL` / `PL_BUILD_VARIANT` / `PL_BUILD_LOCATION` and defaults docker
+image builds to CI-only. A local `build:dev-remote` therefore emitted binary-only
+descriptors, and a block built that way cannot run on a k8s deployment at all —
+that runner launches containers exclusively and rejects a command with no image
+as "docker is not set".
+
+They now use `block-tools software build`, which honours those variables: variant
+`all` builds the images and location `remote` pushes them, the same way
+tcr-disco and the other 23 already-migrated software packages work. `PL_PKG_DEV`
+drops out of the root scripts and `turbo.json`, since `PL_BUILD_LOCATION` carries
+what it used to say.
+
+No change to what the software does — only to how it is built.

package.json

@@ -4,7 +4,7 @@
     "index.js"
   ],
   "scripts": {
-    "test": "env PL_PKG_DEV=local PL_BUILD_CHANNEL=dev PL_BUILD_VARIANT=binary PL_BUILD_LOCATION=local turbo run test --concurrency 1",
+    "test": "env PL_BUILD_CHANNEL=dev PL_BUILD_VARIANT=binary PL_BUILD_LOCATION=local turbo run test --concurrency 1",
     "test:dry-run": "env PL_BUILD_CHANNEL=dev PL_BUILD_VARIANT=binary PL_BUILD_LOCATION=local turbo run test --dry-run=json",
     "mark-stable": "turbo run mark-stable",
     "watch": "turbo watch build",
@@ -15,7 +15,7 @@
     "check": "turbo run check",
     "do-pack": "turbo run do-pack",
     "upgrade-sdk": "block-tools structure refresh --update-deps-only && pnpm i && block-tools structure refresh && pnpm i && pnpm fmt",
-    "build:dev-local": "env PL_PKG_DEV=local PL_BUILD_CHANNEL=dev PL_BUILD_VARIANT=all PL_BUILD_LOCATION=local turbo run build",
+    "build:dev-local": "env PL_BUILD_CHANNEL=dev PL_BUILD_VARIANT=all PL_BUILD_LOCATION=local turbo run build",
     "build:dev-remote": "env PL_BUILD_CHANNEL=dev PL_BUILD_VARIANT=all PL_BUILD_LOCATION=remote turbo run build",
     "build:dev-no-software": "env PL_BUILD_CHANNEL=dev PL_BUILD_VARIANT=none turbo run build",
     "build:dev-binary-existing": "env PL_BUILD_CHANNEL=dev PL_BUILD_USE_PUBLISHED=true turbo run build",

pnpm-workspace.yaml

@@ -28,7 +28,7 @@ catalog:
   # than introduced here, per the spec's requirement that the instrument already exist
   # in the workspace. Versions follow blocks/redefine-clonotypes, its other consumer.
   "@platforma-open/milaboratories.runenv-python-3": ^1.1.16
-  "@platforma-open/milaboratories.software-anarci": ^0.0.3
+  "@platforma-open/milaboratories.software-anarci": ^1.0.1
 
   "vue": 3.5.24
 

software/column-profile/package.json

@@ -2,10 +2,10 @@
   "name": "@platforma-open/milaboratories.import-vdj.column-profile",
   "version": "1.1.0",
   "description": "Profile a csv/tsv's columns: value type and whether they hold amino-acid variable domains",
+  "type": "module",
   "scripts": {
-    "build": "pl-pkg build",
-    "prepublishOnly": "pl-pkg prepublish",
-    "do-pack": "rm -f *.tgz && pl-pkg build && pnpm pack && mv platforma-open*.tgz package.tgz",
+    "build": "block-tools software build",
+    "do-pack": "shx rm -f *.tgz && block-tools software build && pnpm pack && shx mv platforma-open*.tgz package.tgz",
     "changeset": "changeset",
     "version-packages": "changeset version"
   },
@@ -14,8 +14,8 @@
   ],
   "dependencies": {},
   "devDependencies": {
-    "@platforma-sdk/package-builder": "catalog:",
-    "@platforma-open/milaboratories.runenv-python-3": "catalog:"
+    "@platforma-open/milaboratories.runenv-python-3": "catalog:",
+    "@platforma-sdk/block-tools": "catalog:"
   },
   "block-software": {
     "entrypoints": {

software/column-profile/src/main.py

@@ -55,6 +55,23 @@ def value_type(value: str) -> int:
     return T_DOUBLE
 
 
+# The workflow passes a name rather than the character. On the k8s and google-batch runners the
+# backend serialises argv with Go's %q and re-runs it through `sh -c`: a real tab arrives here as
+# the two characters \ and t, which csv.reader rejects outright. A plain word survives that
+# round-trip, so the name is what crosses the boundary and the character is chosen here.
+SEPARATOR_NAMES = {"tab": "\t", "comma": ",", "semicolon": ";"}
+
+
+def resolve_separator(separator: str) -> str:
+    resolved = SEPARATOR_NAMES.get(separator, separator)
+    if len(resolved) != 1:
+        raise SystemExit(
+            f"--separator must be one of {sorted(SEPARATOR_NAMES)} or a single character,"
+            f" got {separator!r}"
+        )
+    return resolved
+
+
 def profile(path: str, separator: str) -> dict:
     with open(path, newline="", encoding="utf-8-sig") as f:
         reader = csv.reader(f, delimiter=separator)
@@ -97,11 +114,15 @@ def profile(path: str, separator: str) -> dict:
 def main() -> None:
     p = argparse.ArgumentParser(description="Profile a csv/tsv's columns")
     p.add_argument("--input", required=True, help="Input csv or tsv")
-    p.add_argument("--separator", required=True, help="Field separator")
+    p.add_argument(
+        "--separator",
+        required=True,
+        help='Field separator: a name ("tab", "comma", "semicolon") or the character itself',
+    )
     p.add_argument("--output", required=True, help="Output JSON")
     args = p.parse_args()
 
-    result = profile(args.input, args.separator)
+    result = profile(args.input, resolve_separator(args.separator))
     with open(args.output, "w") as f:
         json.dump(result, f, sort_keys=True)
 

software/region-annotation/package.json

@@ -2,10 +2,10 @@
   "name": "@platforma-open/milaboratories.import-vdj.region-annotation",
   "version": "1.1.0",
   "description": "Locate FR/CDR region boundaries in amino-acid antibody variable domains from ANARCI numbering",
+  "type": "module",
   "scripts": {
-    "build": "pl-pkg build",
-    "prepublishOnly": "pl-pkg prepublish",
-    "do-pack": "rm -f *.tgz && pl-pkg build && pnpm pack && mv platforma-open*.tgz package.tgz",
+    "build": "block-tools software build",
+    "do-pack": "shx rm -f *.tgz && block-tools software build && pnpm pack && shx mv platforma-open*.tgz package.tgz",
     "changeset": "changeset",
     "version-packages": "changeset version"
   },
@@ -14,8 +14,8 @@
   ],
   "dependencies": {},
   "devDependencies": {
-    "@platforma-sdk/package-builder": "catalog:",
-    "@platforma-open/milaboratories.runenv-python-3": "catalog:"
+    "@platforma-open/milaboratories.runenv-python-3": "catalog:",
+    "@platforma-sdk/block-tools": "catalog:"
   },
   "block-software": {
     "entrypoints": {

software/xlsx-to-csv/package.json

@@ -2,10 +2,10 @@
   "name": "@platforma-open/milaboratories.import-vdj.xlsx-to-csv",
   "version": "1.1.0",
   "description": "Convert the first worksheet of a workbook to CSV so the pipeline only ever sees csv/tsv",
+  "type": "module",
   "scripts": {
-    "build": "pl-pkg build",
-    "prepublishOnly": "pl-pkg prepublish",
-    "do-pack": "rm -f *.tgz && pl-pkg build && pnpm pack && mv platforma-open*.tgz package.tgz",
+    "build": "block-tools software build",
+    "do-pack": "shx rm -f *.tgz && block-tools software build && pnpm pack && shx mv platforma-open*.tgz package.tgz",
     "changeset": "changeset",
     "version-packages": "changeset version"
   },
@@ -14,8 +14,8 @@
   ],
   "dependencies": {},
   "devDependencies": {
-    "@platforma-sdk/package-builder": "catalog:",
-    "@platforma-open/milaboratories.runenv-python-3": "catalog:"
+    "@platforma-open/milaboratories.runenv-python-3": "catalog:",
+    "@platforma-sdk/block-tools": "catalog:"
   },
   "block-software": {
     "entrypoints": {

turbo.json

@@ -14,7 +14,6 @@
       "inputs": ["$TURBO_DEFAULT$"],
       "env": [
         "PL_DOCKER_REGISTRY_PUSH_TO",
-        "PL_PKG_DEV",
         "PL_BUILD_CHANNEL",
         "PL_BUILD_VARIANT",
         "PL_BUILD_LOCATION",

workflow/src/prerun.tpl.tengo

@@ -29,7 +29,12 @@ wf.body(func(args) {
 		// The mapping dropdowns need the file's headers, and on this door nothing upstream
 		// supplies them. The separator comes from what the UI read out of the file's own first
 		// line rather than from the filename, which may not match the content.
-		separator := readable.extension == "csv" ? "," : "\t"
+		//
+		// A name, not the character itself: on the k8s and google-batch runners the backend
+		// serialises argv with Go's %q and re-runs it through `sh -c`, which turns a real tab
+		// into the two characters \t and leaves them that way. Only plain words survive that
+		// round-trip intact. profile-columns maps the name back.
+		separator := readable.extension == "csv" ? "comma" : "tab"
 		// The whole file, not a head: the panel needs each column's value type and whether it
 		// holds amino-acid domains, and both are wrong if taken from a sample. See
 		// profile-columns.

workflow/src/profile-columns.tpl.tengo

@@ -21,6 +21,8 @@ self.defineOutputs("profile")
 
 self.body(func(inputs) {
 	ll.assert(inputs.file != undefined, "profile-columns: file is required")
+	// A name — "tab" or "comma" — not the character. See prerun: a real tab does not survive
+	// the `sh -c` round-trip the k8s runner puts argv through. main.py maps the name back.
 	ll.assert(inputs.separator != undefined, "profile-columns: separator is required")
 
 	cmd := exec.builder().