Use perl58delta to Audit an Old Perl Upgrade

You will turn perl58delta into a short compatibility audit for a Perl 5.6 to 5.8 migration. The page is a release note, not a switch that changes Perl, so the useful result is a written list of code and binary checks to run before an upgrade. Allow about 20 minutes for a small application, plus rebuild time for any XS modules.

The installed page is from the perl-doc package, version 5.38.2-3.2ubuntu0.6, and describes Perl 5.8.0. The local Perl interpreter is 5.38.2. That distinction matters: the notes describe a historical compatibility boundary, while the commands below confirm what the current machine does today.

1. Confirm which documentation you are reading

Start with ordinary, read-only checks. No elevated privileges are needed.

$ command -v perl
/usr/bin/perl
$ perl -v | sed -n '1,4p'
This is perl 5, version 38, subversion 2 (v5.38.2) built for x86_64-linux-gnu-thread-multi
$ dpkg-query -W -f='${Package} ${Version}\n' perl-doc
perl-doc 5.38.2-3.2ubuntu0.6
$ man perl58delta

perl58delta takes no operational arguments in this workflow. Read it as a record of changes between Perl 5.6.0 and 5.8.0. Changes marked [561] were already present in the 5.6.1 maintenance release, and [561+] means that later 5.6.1 work was developed further for 5.8.0.

Checkpoint: if your application is not moving across this boundary, stop. A release note can still explain old code, but it is not a substitute for the delta document matching your actual source and target versions.

2. Find the binary compatibility boundary

The first migration decision is whether the application has compiled Perl extensions. Perl 5.8 is not binary-compatible with earlier Perl releases, and the manpage specifically requires XS modules to be recompiled. Pure Perl modules should normally continue to work, but test them rather than treating that sentence as a guarantee.

$ perl -MConfig -e 'print "perl=$^V\narch=$Config{archname}\nsitearch=$Config{sitearch}\n"'
perl=v5.38.2
arch=x86_64-linux-gnu-thread-multi
sitearch=/usr/local/lib/x86_64-linux-gnu/perl/5.38.2

Inventory modules that contain compiled code, including application-local extensions and modules loaded by deployment tools. Rebuild them with the target interpreter, then run the application's test suite. On 64-bit platforms the release notes also describe allocator changes, so do not copy an old compiled module tree into the new Perl installation.

Warning: this is the point at which a blind shared-library replacement can fail. Keep the old runtime and module tree available until the rebuilt application has passed its tests. Recovery is a deployment rollback to that known-good pair, not a Perl command.

3. Test the PerlIO and encoding assumptions

PerlIO became the default I/O implementation in 5.8. It introduced layers such as :raw, :utf8 and :encoding(). The practical trap is treating a binary stream as text, or assuming that disabling CRLF translation alone makes a stream byte-for-byte raw.

$ perl -MPerlIO -e 'print join("\n", PerlIO::get_layers(*STDOUT)), "\n"'
unix
perlio

The exact layer list is platform-dependent, so use this as an observation rather than a hard-coded expected output. Test the application's actual files. Text input should name its encoding deliberately where the data contract requires it:

open my $input, '<:encoding(UTF-8)', $path
    or die "open $path: $!";
open my $binary, '<:raw', $image
    or die "open $image: $!";

Do not add use utf8 as a general Unicode switch. In the 5.8 model, Unicode status belongs mainly to the data; the pragma remains relevant when the Perl source file itself is written in UTF-8. Check source encoding, input decoding and binary handling as separate concerns.

4. Exercise Unicode and sorting changes

The release notes cover Unicode properties, scripts, regular expressions and I/O, and they say that glob() now sorts filenames alphabetically by default. These are good focused regression tests because they expose assumptions without modifying the system.

$ perl -e 'my $word = "café"; print $word =~ /\p{Letter}+/u ? "unicode-match\n" : "no-match\n"'
unicode-match
$ tmpdir=$(mktemp -d)
$ trap 'rm -rf "$tmpdir"' EXIT
$ touch "$tmpdir/z" "$tmpdir/a"
$ TMPDIR="$tmpdir" perl -e 'print join(" ", glob "$ENV{TMPDIR}/*"), "\n"'
/tmp/tmp.example/a /tmp/tmp.example/z

The temporary directory name will differ. The important check is that the final two path names are in alphabetical order. The cleanup trap removes only the temporary directory created by this test. If an application depends on filesystem order, replace that assumption with an explicit sort and test the expected locale behaviour.

5. Review source-level traps before testing services

Several 5.8 changes can alter code without changing its command line. The release notes call out the new interpreter-thread model, safe signals, integer-preserving arithmetic, better prototype checks and the fact that arrays interpolate in double-quoted strings. They also list removed or deprecated constructs, including the old Thread model, uppercase comparison aliases, tr///C and tr///U, pseudo-hash internals, and unqualified dump().

$ perl -we 'my @names = ("Ada", "Grace"); print "@names\n"'
Ada Grace
$ perl -we 'print "[email protected]\n"'
fred.com at -e line 1.

The second command demonstrates the warning-prone email-address trap: an unescaped @ in a double-quoted string is treated as array interpolation. Write fred\@example.com or use a single-quoted string when that is the intended literal text. Run your test suite with warnings enabled, and search source for the deprecated forms named in the page before you change the interpreter.

6. Record the migration decision

Do not convert a release-note list into an automatic edit. Record each item as one of three outcomes: tested and unchanged, changed and covered by a regression test, or not applicable with a reason. For service deployments, perform the final test under the same user, environment and input files as production. No command in this guide needs sudo; installing or rebuilding packages belongs to your normal change process.