Phase 4: -o textmode RMS record decoder

src/recfmt.c implements the on-the-fly text view of an ODS-2 file:

  STMLF/STM   pass-through
  STMCR       CR -> LF substitution
  VAR         strip the 16-bit length prefix, walk the body, append LF
              when FAB$M_CR is set, pad to even-byte boundaries
  VFC         like VAR but skip the fat$b_vfcsize-byte fixed header
  FIX         emit fixed-size records, append LF when FAB$M_CR is set
  unknown     fall back to a raw byte pass-through

FAB$M_BLK (records do not span block boundaries) and the 0xffff
end-of-block padding marker are honoured, mirroring how a VMS reader
would treat the file.

The decoder produces the entire decoded stream into a malloc'd buffer
the first time it's asked, keyed by FID in a small linked-list cache.
ops_getattr returns the cached length as st_size; ops_read serves
slices out of the cached buffer.  recfmt_clear_cache() runs on
dismount so a transient mount doesn't leak memory.

Bump actions/checkout and actions/upload-artifact to v5 to drop the
Node.js 20 deprecation warning.

Author: mancausoft

.github/workflows/build.yml

@@ -7,11 +7,6 @@ on:
     branches: [ main ]
   workflow_dispatch:
 
-env:
-  # Opt JS-based actions into Node.js 24 ahead of the default flip
-  # (Node 20 is being deprecated on GitHub-hosted runners).
-  FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: "true"
-
 jobs:
   debian:
     runs-on: ubuntu-latest
@@ -24,7 +19,7 @@ jobs:
           apt-get install -y --no-install-recommends \
             build-essential pkg-config libfuse3-dev fuse3 git ca-certificates
 
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v5
 
       - name: Show toolchain
         run: |
@@ -46,7 +41,7 @@ jobs:
         run: ./fuse-ods2 -h || true
 
       - name: Upload binary
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@v5
         with:
           name: fuse-ods2-debian
           path: fuse-ods2

Makefile

@@ -51,7 +51,8 @@ LIB_OBJS := $(LIB_SRCS:%.c=$(BUILD)/%.o)
 NO_FUSE_SRCS := \
     src/compat_glue.c \
     src/phyfuse.c     \
-    src/lookup.c
+    src/lookup.c      \
+    src/recfmt.c
 
 FUSE_SRCS := \
     src/fuse_ods2.c   \

TODO.md

@@ -65,11 +65,12 @@ brew install macfuse pkg-config
 - [ ] Honour highwater (return zeros past `fcb->highwater`)  *(deferred: most images are dense)*
 - [ ] Test: `cmp` STREAM_LF and FIX files against an ods2 extraction  *(deferred to Linux validation)*
 
-## Phase 4 - text-mode (`-o textmode`)
-- [ ] `src/recfmt.c`: streaming reader (FCB, byte_offset, len) -> "decoded" bytes
-- [ ] Per-record-format strategies (FIX/VAR/VFC/STM/STMLF/STMCR)
-- [ ] Recompute `st_size` in `getattr` when textmode (cache the logical length)
-- [ ] Test on `.LIS` / `.TXT` files in VAR format
+## Phase 4 - text-mode (`-o textmode`) [CODE DONE, RUNTIME TBD]
+- [x] `src/recfmt.c`: full-file decoder with per-FID buffer cache
+- [x] Per-record-format strategies (FIX/VAR/VFC/STM/STMLF/STMCR + UDF passthrough)
+- [x] Recompute `st_size` in `getattr` when textmode (cache the logical length)
+- [x] Cache cleared on dismount
+- [ ] Test on `.LIS` / `.TXT` files in VAR format  *(deferred to phase 7 e2e)*
 
 ## Phase 5 - offset + volume set
 - [ ] `-o offset=N`: implemented in phyfuse, exposed on the CLI

src/fuse_ods2.c

@@ -16,6 +16,7 @@
 #include <unistd.h>
 
 #include "ods2_ops.h"
+#include "recfmt.h"
 
 #include "access.h"
 #include "f11def.h"
@@ -216,6 +217,7 @@ static int do_mount( void ) {
 }
 
 static void do_dismount( void ) {
+    recfmt_clear_cache();
     if( ods2_vcb != NULL ) {
         dismount( ods2_vcb, 0 );
         ods2_vcb = NULL;

src/ops.c

@@ -16,6 +16,7 @@
 #include <unistd.h>
 
 #include "ods2_ops.h"
+#include "recfmt.h"
 
 #include "access.h"
 #include "f11def.h"
@@ -74,6 +75,24 @@ int ods2_getattr( const char *path, struct stat *st,
     if( !ods2_rt.force_gid && st->st_gid == 0 )
         st->st_gid = getgid();
 
+    /* In textmode the on-disk size is wrong (record headers, padding,
+     * implicit LFs).  Open the file briefly to compute the decoded
+     * length; the result is cached, so subsequent stat/read are cheap.
+     */
+    if( ods2_rt.textmode && S_ISREG( st->st_mode )
+        && recfmt_textmode_eligible( head ) ) {
+        deaccesshead( vioc, NULL, 0 );
+
+        struct FCB *fcb = NULL;
+        if( $SUCCESSFUL( accessfile( ods2_vcb, &fid, &fcb, 0 ) ) ) {
+            ssize_t lsize = recfmt_logical_size( fcb );
+            if( lsize >= 0 )
+                st->st_size = (off_t)lsize;
+            deaccessfile( fcb );
+        }
+        return 0;
+    }
+
     deaccesshead( vioc, NULL, 0 );
     return 0;
 }
@@ -225,6 +244,15 @@ int ods2_read( const char *path, char *out, size_t size, off_t offset,
     if( fcb == NULL )
         return -EBADF;
 
+    /* In textmode redirect through the record-format decoder; it has
+     * its own size accounting and cached buffer. */
+    if( ods2_rt.textmode && fcb->head != NULL
+        && recfmt_textmode_eligible( fcb->head ) ) {
+        ssize_t n = recfmt_read( fcb, offset, out, size );
+        if( n < 0 ) return -EIO;
+        return (int)n;
+    }
+
     /* Logical end of file derived from the RECATTR.  EOF block is the
      * first block past the last block holding data; ffb is the count
      * of bytes used in the last block.  When efblk == 0 the file has