cloc ignores nested .tar.gz inside .src.rpm

[Tenount]

Describe the bug

When running cloc directly on a source RPM file (.src.rpm), it reports zero code lines detected. The .src.rpm contains a nested .tar.gz archive with C++ sources, but cloc does not recursively extract the inner archive — it only processes the top-level files (cppzmq.spec, cppzmq-4.10.0.tar.gz), leaving the C++ code inside the tarball uncounted.

After manual extraction (rpm2cpio | cpio | tar), the same files are correctly counted by cloc.

cloc; OS; OS version

• cloc version: 2.00
• OS: Alpine Linux
• OS version: 3.20 (aarch64)
• Reproducible via podman or docker

To Reproduce

1. Start an Alpine container:

   podman run --rm -it alpine:3.20 /bin/sh

2. Install dependencies:

   apk add --no-cache cloc gzip rpm2cpio cpio tar wget

3. Download a Fedora source RPM (cppzmq — a pure C++ library):

   wget -q \
     "https://dl.fedoraproject.org/pub/fedora/linux/releases/44/Everything/source/tree/Packages/c/cppzmq-4.10.0-12.fc44.src.rpm" \
     -O cppzmq.src.rpm

4. Run cloc directly on the .src.rpm file:

   cloc cppzmq.src.rpm

5. Observe the result — only 1 text file counted, C++ inside the nested .tar.gz is ignored:

   104 blocks
           1 text file.
           0 unique files.
           2 files ignored.
   

   The .src.rpm contains cppzmq.spec (the 1 text file) and cppzmq-4.10.0.tar.gz (ignored — cloc does not recursively extract it).

6. Now extract manually and run cloc again:

   mkdir ext && rpm2cpio cppzmq.src.rpm | cpio -idm -D ext 2>/dev/null
   tar xzf ext/cppzmq-4.10.0.tar.gz -C ext/
   cloc ext/cppzmq-4.10.0/

7. Observe the correct result:

   github.com/AlDanial/cloc v 2.00  T=0.02 s (1625.4 files/s, 403155.3 lines/s)
   -------------------------------------------------------------------------------
   Language                     files          blank        comment           code
   -------------------------------------------------------------------------------
   C/C++ Header                     3            428            258           2883
   C++                             17            421             96           2832
   CMake                            7             59             27            220
   YAML                             2             17             12            219
   Markdown                         1             34              0            162
   Bourne Shell                     1              1              4             16
   -------------------------------------------------------------------------------
   SUM:                            31            960            397           6332
   -------------------------------------------------------------------------------
   

Expected result

When running cloc cppzmq.src.rpm, the language table should match the extracted content — showing C++ (~2832 lines) and C/C++ Header (~2883 lines). Instead, cloc silently returns success with no languages counted.

Actual result

$ cloc cppzmq.src.rpm
104 blocks
       1 text file.
       0 unique files.
       2 files ignored.

Only cppzmq.spec is counted; cppzmq-4.10.0.tar.gz (containing all the C++ source) is silently skipped.

Minimal reproducible Dockerfile

FROM alpine:3.20

RUN apk add --no-cache cloc gzip rpm2cpio cpio tar wget

RUN wget -q \
  "https://dl.fedoraproject.org/pub/fedora/linux/releases/44/Everything/source/tree/Packages/c/cppzmq-4.10.0-12.fc44.src.rpm" \
  -O /cppzmq.src.rpm

RUN mkdir -p /extracted && \
  rpm2cpio /cppzmq.src.rpm | cpio -idm -D /extracted 2>/dev/null && \
  tar xzf /extracted/cppzmq-4.10.0.tar.gz -C /extracted/

RUN echo "=== TEST 1: cloc on .src.rpm directly ===" && \
  cloc /cppzmq.src.rpm

RUN echo "=== TEST 2: cloc after extraction ===" && \
  cloc /extracted/cppzmq-4.10.0/

Additional context

• The .src.rpm contains: cppzmq-4.10.0.tar.gz + cppzmq.spec
• cloc correctly identifies and extracts the .src.rpm (exit code 0, rpm2cpio runs)
• The extracted tarball (cppzmq-4.10.0.tar.gz) inside the RPM is not recursively processed
• Related: cloc should try figuring out more than one level-deep archives  #161 (same root cause — cloc does not recursively extract nested archives, e.g. .tar.gz inside .deb; closed in 2017 without a fix)

-v 3 verbose trace for cloc cppzmq.src.rpm

-> load_from_config_file(/root/.config/cloc/options.txt)
<- load_from_config_file() (no such file: /root/.config/cloc/options.txt)
-> replace_git_hash_with_tarfile(cppzmq.src.rpm)
-> no_autogen(0)
<- no_autogen()
-> uncompress_archive_cmd(/cppzmq.src.rpm)
<- uncompress_archive_cmd
mkdir /tmp/CEQfe4j5be
cd    /tmp/CEQfe4j5be
rpm2cpio '/cppzmq.src.rpm' | cpio -i
104 blocks
-> make_file_list(/tmp/CEQfe4j5be)
Using temp file list [/tmp/01ryDCKuFb]
-> find_preprocessor(/tmp/CEQfe4j5be)
<- find_preprocessor(cppzmq-4.10.0.tar.gz cppzmq.spec)
       1 text file.
classifying /tmp/CEQfe4j5be/cppzmq.spec
-> classify_file(/tmp/CEQfe4j5be/cppzmq.spec)
/tmp/CEQfe4j5be/cppzmq.spec extension=[spec]
-> peek_at_first_line(/tmp/CEQfe4j5be/cppzmq.spec)
-> first_line(/tmp/CEQfe4j5be/cppzmq.spec, 1)
<- first_line(/tmp/CEQfe4j5be/cppzmq.spec, 1, '## START: Set by rpmautospec')
<- peek_at_first_line(/tmp/CEQfe4j5be/cppzmq.spec)
<- classify_file(/tmp/CEQfe4j5be/cppzmq.spec)=(unknown)
<- make_file_list()
-> remove_duplicate_files
<- remove_duplicate_files
       0 unique files.
-> count_files()
<- count_files()
       2 files ignored.
-> write_null_results
<- write_null_results

Key observation: find_preprocessor() returns cppzmq-4.10.0.tar.gz and cppzmq.spec, but cloc never recurses into the .tar.gz — it only classifies cppzmq.spec (as unknown) and writes null results.