| Age | Commit message (Collapse) | Author |
|---|
|
A real album (irregular.ninja/so.war.das) surfaced the bug: a Fujifilm
X100V whose battery died reset its clock to 2000-01-01, and every EXIF
timestamp on that roll was stamped with the bogus date instead of being
omitted. Trusting it at face value sorted that whole roll to the front
of the album, ahead of hundreds of correctly-dated 2020s photos.
EXIF years before 2001 are now treated the same as a missing date tag.
The no-date fallback also switches from source mtime to filename order,
since mtime is not reliable either -- bulk-copying/rsyncing an incoming
directory commonly rewrites every file's mtime to the transfer time,
unrelated to capture order -- while sequential camera filenames track
real shooting order even across a clock reset.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
|
|
Adds a new yes/no config setting (default no, preserving current behavior)
that orders the main album's photos by EXIF date taken (ascending) instead
of the default filename/shuffle order. Reuses the existing EXIF cache and
tag fallback chain (DateTimeOriginal -> DateTimeDigitized -> DateTime)
already used for tooltips/details/stats, so ordering never disagrees with
what those features show. Photos with no usable EXIF date fall back to
their source file's mtime, staying fully deterministic and crash-free.
CHRONOLOGICAL_ORDER takes precedence over SHUFFLE when both are enabled,
documented in album-photo-select.source.sh and docs/configuration.md.
Wired through the config registry (CONFIG_SPECS), validation, CLI flags
(--chronological/--no-chronological), --print-config, --dry-run,
--verbose logging, and shuriken.json generation metadata. Adds unit and
end-to-end tests covering default-off behavior, EXIF-date ordering with
shuffle precedence, and mtime fallback for EXIF-less photos.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
|
|
shuriken relied on GNU-only extensions (find -printf, stat -c, cp -a,
sort -R) and hard-refused to run outside Linux. Resolve each of the four
tools once at startup to a FIND/STAT/CP/SORT variable, preferring a
g-prefixed sibling (gfind, gstat, gcp, gsort) when present on PATH -
mirroring the $SED/$GREP/$DATE tool-selection pattern in the sibling
gemtexter project - so the same source runs unmodified on Linux, macOS,
and FreeBSD once GNU coreutils/findutils are installed there.
- src/lib/compat.source.sh: add resolve_gnu_tool + FIND/STAT/CP/SORT;
add a fast verify_gnu_tool_versions preflight (--version contains
"GNU") ahead of the existing behavioral probes, which now run against
the resolved variables instead of hardcoded tool names; split the
probes and error reporting into their own ~30-line functions.
- Replace the GNU-only find -printf / stat -c / cp -a / sort -R call
sites in photo-list, image, album-photo-select, metadata-cache,
config.staging, and random source files with the resolved variables.
POSIX-portable find/sort calls elsewhere are untouched.
- README.md: drop the "Linux-only" claim; document brew/pkg GNU
coreutils+findutils install steps for macOS and FreeBSD.
- tests/cli.sh: add coverage for the g-prefixed-sibling preference and
for the new --version-string preflight's error message, alongside the
existing behavioral-probe guard tests.
just build / just check-generated / just shellcheck / just test /
git diff --check all pass.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
|
|
Album owners can now set DETAILS_PAGE=no (or pass --no-details) to skip
generating each photo's *-details.html EXIF summary page and its "Details"
navigation redirects, without touching the normal thumbnail overview,
per-photo view pages, EXIF tooltips, or STATS_PAGE, which all stay
independently controlled. Every "Details" link (on view pages and stats
filter mini-album view pages) and every "-details" redirect stub is gated
on the setting so no generated page ever links to a file that was not
rendered. Wired the new field through CONFIG_SPECS (registry-driven
defaults/validation/print-config/CLI override), the --details/--no-details
CLI flags, usage() help, the verbose effective-config log, shuriken.json
generation metadata, and the dry-run plan. DETAILS_PAGE=yes (the default)
keeps prior output byte-for-byte identical.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
|
|
Closes the last output-coverage gap from the gauge-bar audit: the flush
guarantee (a page's grid-cell total is a multiple of 12 so it forms a
complete rectangle at 2/3/4/6 columns) was only unit-tested on the
alignment helper and via markup presence. The dense-packing check that
verified it on real albums was a throwaway. test_generate_preview_grid_
emits_every_tile_shape now computes the generated page-1 cell total
(features=4 cells, others=1) and asserts it is a positive multiple of 12,
and that the pinned seed still yields both a feature and a subdivided tile.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
Closes the test-coverage blind spot the gauge-bar regression exposed: the
rendered grid tile types had zero content assertions. Adds three
deterministic integration tests (seeded layouts):
- test_generate_preview_grid_emits_every_tile_shape: single thumbnail
(class='thumb'), 2x2 feature (class='feature' + a.feature span CSS),
subdivided (<div class='tile'> + subthumb), the thumbs-grid container,
all four breakpoint column rules (repeat 2/3/4/6), the view-page body
and prev/next nav arrows. Pinned 30 photos, RANDOM_SEED, FEATURE=60,
SUBDIVIDE=70, SHUFFLE=no for reproducibility.
- test_generate_short_final_page_emits_fill_row_tiles: the short-final-page
full-width banner (class='fill-row' + grid-column:1/-1), absent on a full
page. Pinned 13 photos, MAXPREVIEWS=10.
- test_generate_splash_page_renders_enter_link_and_photo: splash <main>,
title, splash photo, Enter-album link.
Principle: every generated output format now has >=1 assertion on its
meaningful content, not just existence or render-equality. No src changes.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
Regression from or0 (STATS_* accessors): stats_category_max compared
values via `(( _stats_counts_ref[key] > max ))`. For a nameref to an
ASSOCIATIVE array, an arithmetic subscript is itself arithmetic-evaluated,
so a string key (e.g. "sony") resolves to an unset variable -> 0, i.e. it
always read index 0 and returned max 0. Every bar then scaled to
width:0% (counts/percentages were unaffected, so only the bars looked
broken). Read the value through a quoted ${assoc[$key]} expansion first,
then compare.
The byte-identical double-render test missed this because both renders
were equally broken; added explicit non-zero bar-width assertions
(100% / 50%) to test_render_stats_page_renders_sections_and_escapes.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
Introduce src/lib/config.spec.source.sh: CONFIG_SPECS, a single '|'-delimited
config-field registry (name|default|has_default|cli_overridable|validation|
print_kind), using the same spec idiom as ACTION_SPECS / TEMPLATE_RENDER_FIELD_SPECS.
This replaces the parallel, hand-maintained config-knowledge lists that caused the
TARBALL_INCLUDE default-drift bug (fixed in 7r0).
Derived consumers (behaviour byte-identical):
- apply_config_defaults: loops the registry applying VAR="${VAR:-default}" for
has_default=yes scalars; arrays keep their declare -p guards.
- CLI_CONFIG_OVERRIDE_TARGETS: built from cli_overridable=yes (verified to match
CLI_OPTION_SPEC's config= targets exactly).
- print_config: emits in registry order dispatching on print_kind.
- validate_common_config: required set + per-field rule come from the registry via
config_spec_validation + validate_config_field_required/_kind; two-phase order and
historical reporting order preserved.
log_configured_action and the dry-run plan are not yet converted (bespoke prose /
intermixed computed values). shellcheck: TARBALL_SUFFIX lost its visible literal
assignment, annotated at the archive use site.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
The EXIF cache dir ($(dirname "$DIST_DIR")/cache/exif) was recomputed inline,
byte for byte, in both cached_photo_identify_output (read/write) and
clear_exif_cache (--force/--clean removal). The plain parent of DIST_DIR
(dirname "$DIST_DIR") was likewise recomputed in metadata-cache and in
action.source.sh's staging-artifact cleanup.
Extract two helpers computed once from DIST_DIR:
- working_dir() in config.source.sh (next to existing_parent_dir, the other
DIST_DIR-parent resolver): plain `dirname "$DIST_DIR"`.
- exif_cache_dir() in metadata-cache.source.sh (owns the EXIF cache):
`working_dir()/cache/exif`.
Route cached_photo_identify_output, clear_exif_cache, and
clean_generation_staging_artifacts through them so the cache reader and the
cleaner can never drift to different directories. Paths are byte-identical to
the prior inline code (dirname semantics, cache/exif suffix, trailing-slash and
relative/absolute handling all preserved). Stale "recompute the cache dir"
comments removed; helpers document the path once. Add a unit test asserting the
helpers agree and resolve beside dist for several DIST_DIR shapes.
Broader DIST_DIR parameterization of leaf pipeline helpers was intentionally
left out of scope (only the duplicated path computation is centralized here).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
Wrap each rsync in sync_dist in run_with_timeout (new SYNC_TIMEOUT config,
default 300s) so a hung or unreachable mirror cannot block the whole sync,
matching every other external call. Make destinations isolated: under
set -euo pipefail a single failing destination used to abort the loop and
silently skip the rest. Now each destination runs under a localized set +e
(the project's refresh_splash idiom), results are collected per destination,
a clear pass/fail summary is logged, and sync returns non-zero if any
destination failed while still attempting all of them.
SYNC_TIMEOUT is plumbed like TAR_TIMEOUT: shuriken.default.conf,
apply_config_defaults, print_config, verbose config log, and positive-integer
validation in both validate_config (generate path) and validate_sync_config
(sync path). No CLI flag, matching TAR_TIMEOUT.
Tests: a sync where one destination fails still attempts the others and exits
non-zero with the summary; SYNC_TIMEOUT=0 is rejected as a positive integer.
Adds install_rsync_spy_failing_one helper and updates the print_config
expected blocks.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
Running `bash src/shuriken.sh --generate ...` directly from a source
checkout printed "command not found" for camera_label_from_make_model,
photo_exif_values_to and cached_photo_identify_output for every photo,
silently emptying EXIF tooltips/details and, with STATS_PAGE=yes,
omitting the whole stats/ tree. The hand-maintained source list inside
the SHURIKEN_LIB_SOURCES_BEGIN/END marker block had drifted from the
authoritative Justfile LIB_SOURCES, missing metadata-label,
metadata-cache, stats-aggregate, stats-render and stats-filter-album.
`just build` replaces the marker block with LIB_SOURCES when generating
bin/shuriken, so the installed binary and the bin-based test suite never
noticed; only direct src execution was affected.
- Add the 5 missing `source` lines to the marker block in the same order
as Justfile LIB_SOURCES, so the two lists now match exactly.
- Add tests/cli.sh case test_lib_sources_match_justfile_lib_sources that
extracts the marker-block module names and asserts they equal the
Justfile LIB_SOURCES (same set and order) to prevent future drift.
- shellcheck --check-sourced now follows the 5 newly-sourced libs;
suppress the cross-module nameref false positives (SC2178/SC2128/
SC2154) with explained directives and genuinely fix SC2004
(counts_ref[$key] -> [key]) and quote the TITLE default (${TITLE:-}).
bin/shuriken changes only by these propagated lib edits; the marker-block
source list it generates is unchanged.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
shuriken shells out to GNU-only features of the standard Unix tools
(find -printf, stat -c, cp -a, sort -R). Add require_gnu_tools in a new
src/lib/compat.source.sh, sourced early and invoked from main() before any
action runs. On invocation it feature-probes each tool in a throwaway temp
dir; if any probe fails it prints a clear error naming the offending tool
and exits 1, so non-GNU (macOS/BSD) environments fail fast instead of
producing broken output.
README gains a Platform compatibility section and the requirements line now
mentions GNU coreutils/findutils. Tests cover the find and stat rejection
paths; the shared test helper that builds a coreutils-without-imagemagick
PATH now includes cp and stat (which the guard probes).
|
|
The overview grid used auto-fill columns (an unpredictable count at view
time) while tiles per page were fixed at generation, so the last row was
ragged -- an empty, cut-off bottom-right corner, made worse by a 2x2
feature tile.
CSS (header.tmpl): replace auto-fill with a FIXED column count per width
breakpoint -- 2 (phone) / 3 / 4 / 6 -- all divisors of 12. THUMBHEIGHT no
longer drives the grid (it only sizes the thumbnail files), so its
obsolete render-var is dropped (template.source.sh).
Generator (album-tile-layout, album-thumbnail-html): append_preview_grid
now decides a page's tiles, then snaps the grid-cell total onto a multiple
of 12 before emitting, via two photo-preserving levers --
_grid_split_subdivides_to_add (round up: split subdivided tiles into
singles; preferred, abundant) and _grid_merge_singles_to_remove (round
down: merge adjacent singles). Because 2/3/4/6 all divide 12, a
multiple-of-12 page tiles into a COMPLETE rectangle at every breakpoint:
a flush last row at any window width, with no image distortion
(object-fit: cover). Per-photo preview numbers and all navigation
redirects are unchanged. Tiny pages (a short final page or small stats
mini-album) are left as-is.
Decrements use assignment, not bare "(( --k ))": under set -euo pipefail
an arithmetic command evaluating to 0 returns status 1 and would abort
generate. Helper namerefs are uniquely named and arrays are forwarded by
name to avoid bash circular-nameref errors.
Tests: add test_album_grid_cells_align_to_multiple_of_12 (both levers);
update the two tests that pinned the old auto-fill CSS / render-var list.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
Make the template.source.sh escape/date helper family follow one
consistent shape: each escaper now has a nameref <name>_to form (hot
path, writes a named variable) plus a thin printf wrapper <name> that
delegates to it. The leading "_" now exclusively marks private
helpers; the public escape API (called from sibling modules) is
unprefixed.
- current_date_text now delegates to current_date_text_to so both
forms share the SHURIKEN_CURRENT_DATE_TEXT cache; the printf form no
longer silently re-execs `date` on every direct call. Output
unchanged.
- Drop the misleading "_" prefix on the public escape API and update
all callers: _html_escape->html_escape, _css_string_escape->
css_string_escape, _json_string->json_string, _json_bool->json_bool,
_json_string_escape->json_string_escape.
- Add the missing JSON nameref forms: json_string_escape_to,
json_string_to, json_bool_to (printf wrappers delegate to them).
- Add tests: JSON printf-vs-nameref parity and a current_date_text
caching + nameref-parity check.
No escaping/encoding or date output changes -- API-shape/perf only.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
apply_config_defaults fell back to TARBALL_INCLUDE=no while the user-facing
src/shuriken.default.conf documents (and --init writes) TARBALL_INCLUDE=yes.
The two disagreed, so an album generated from a config that omitted the key
silently dropped the tarball despite the documented default saying otherwise.
'yes' is the authoritative default: the original definition was
`declare -r TARBALL_INCLUDE=yes` (tarball inclusion was on from the start);
the ':-no' fallback was introduced later during a refactor and drifted from
the documented intent. Align apply_config_defaults to ':-yes' and add a
comment explaining the invariant and the history.
Tests: add test_tarball_include_default_matches_init_config asserting the
effective default agrees between a fresh --init config and apply_config_defaults
(and is 'yes'); update the omitted-runtime-defaults --print-config expectation
to the corrected 'yes'. bin/shuriken regenerated via just build.
This is the scoped-down drift fix only. The full CONFIG_SPECS registry plus
cfg_get/SHURIKEN_CFG access-layer rewrite proposed in 7r0 remains DEFERRED.
just test, just shellcheck, just check-generated and git diff --check all pass.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
The `identify -verbose` EXIF-line regex and its array-fill loop were
duplicated in three places that had already drifted: album-metadata's
photo_exif_details_html and _photo_exif_values_to (exif: only) and
stats-aggregate's _stats_parse_identify_stream (exif: plus a native
Geometry -> __geometry path).
Promote one canonical parser, photo_exif_values_to, into
metadata-cache.source.sh next to its sibling cache primitive
cached_photo_identify_output. It reads an identify stream from stdin and
fills a nameref associative array; it is a strict superset of all three
former sites (bare exif: tag keys plus the synthetic __geometry key).
- stats accumulate_photo_stats now calls photo_exif_values_to (stdin);
_stats_parse_identify_stream is removed.
- album _photo_exif_values_to is a thin wrapper that pipes
cached_photo_identify_output through the canonical parser.
- album photo_exif_details_html consumes the same parser and skips the
__geometry key it does not display.
metadata-cache is sourced before both consumers in LIB_SOURCES, so the
canonical parser is available at use time. Regenerated bin/shuriken via
`just build`. Added test_shared_identify_parser_returns_exif_and_geometry
asserting one parse yields both an exif: key and __geometry.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
init_config rewrote the TEMPLATE_DIR line via awk as a bare, unquoted literal
when running --init from a source checkout. A source-root path containing a
space (or any shell metacharacter) produced a config that aborts on source
under "set -euo pipefail": the value was word-split, truncating the path and
running its tail as a command (exit 127), leaving TEMPLATE_DIR empty and
breaking every subsequent action.
Emit the rewritten value as a single-quoted assignment, escaping any embedded
single quote as '\'', so the generated shuriken.conf round-trips through
sourcing regardless of spaces or single quotes. The non-rewrite path (installed
/etc/default/shuriken) is unchanged. Regenerated bin/shuriken via just build.
Tests: test_init and test_init_with_hash_in_source_path now assert the SOURCED
TEMPLATE_DIR (the real contract) instead of the literal line, and a new
test_init_with_space_in_source_path covers the spaced-path regression.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
Pure CSS/HTML, no JavaScript:
- Add a <meta name="viewport"> tag to the two display heads (header.tmpl
and splash.tmpl) so phones stop rendering at a zoomed-out desktop
width.
- Make the overview grid responsive: minmax(min(THUMBHEIGHT, 100%), 1fr)
so a column never overflows a narrow screen, plus a max-width:700px
media query that shows exactly two square columns on phones (2x2
feature tiles become full-width heroes). aspect-ratio keeps cells
square at every width.
- Add overflow-x:hidden to body so the slam/glitch entry animations
(translateX +/-80vw) cannot make a phone scroll sideways.
- Make the menus touch friendly on phones: the navigator/footer/splash
links become large rounded buttons (~44px tap targets) and the "|"
separators (now wrapped in <span class="nav-sep">) are hidden on
mobile while still showing on desktop.
- Fix a pre-existing W3C error: the stats filter view-page image was
missing an alt attribute (and carried an obsolete border attribute).
Verified with headless screenshots at phone width and the W3C Nu HTML
checker + CSS validator (zero errors).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
The overview grid used fixed-width centred columns, leaving large empty
margins left and right on wide screens. Switch it to
repeat(auto-fill, minmax(THUMBHEIGHT, 1fr)) so the columns grow to fill
the full window width (THUMBHEIGHT becomes the minimum cell width), with
aspect-ratio:1 keeping every cell square so the 2x2 feature and
subdivided tiles stay square. Also drop the browser's default body margin
so the space above the header and below the footer matches the small gaps
between the header, grid and footer.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
The per-camera and per-filter stats mini-album galleries rendered their
own plain img.thumb grid, so they did not get the dynamic tiles. Make
them reuse the album's shared grid builder instead.
Generalize append_preview_grid (and build_tile_block /
build_subdivided_tile / build_preview_thumbnail) to take an href_prefix
rather than a page number: the main album passes "<page_num>-" (view
pages "<page>-<n>.html") and the stats mini-albums pass "" (view pages
bare "<n>.html") -- the only difference between the two grids. The
duplicate _stats_filter_thumbnail is removed; _stats_build_filter_thumbs
now calls append_preview_grid, and camera.tmpl wraps the result in the
same thumbs-grid container. Stats galleries now get the identical 2x2
feature tiles, subdivided sub-thumbnails, entry animations and hover
effects as the main overview.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
Allow a single photo to be blown up into a large "feature" tile that
spans a 2x2 block of the album overview, controlled by a new
THUMB_FEATURE_PERCENT (0-100, default 10; 0 disables). Each tile rolls
for a feature first, then for a subdivision, otherwise stays a normal
square.
To pack mixed-size tiles (normal 1x1, subdivided 1x1, feature 2x2)
without gaps, the overview is now a real CSS grid with
grid-auto-flow: dense, so smaller tiles backfill the holes a 2x2 feature
would leave. Tile spacing moved from per-image padding to the grid gap.
Feature tiles reuse the img.thumb class and its dramatic hover.
THUMB_FEATURE_PERCENT is wired through the same layers as
THUMB_SUBDIVIDE_PERCENT: config defaults, 0..100 validation,
--print-config, the --feature CLI flag, the usage text, and the
shuriken.json / --dry-run metadata, with docs and tests updated.
Generated HTML and CSS pass the W3C Nu HTML checker and CSS validator.
Setting both percentages to 0 reproduces the previous all-1x1 grid.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
Make the album preview grid livelier: with a configurable probability
(THUMB_SUBDIVIDE_PERCENT, default 30%) a square thumbnail tile is
subdivided into several smaller thumbnails packed into the same square
footprint, chosen at random from:
- quad: 2x2 squares (4 photos)
- two-wide: two stacked full-width strips (2 photos)
- squares+wide: two squares plus one full-width strip, strip on the
top or the bottom (3 photos)
Each sub-thumbnail stays its own clickable photo with its own view page;
subdivision only groups consecutive photos visually, so preview
numbering and the view/details/redirect pages are unchanged. No new
images are generated (CSS object-fit crops the existing aspect-correct
thumbs into squares or wide strips). Sub-thumbnails get the same random
entry animation and the same dramatic hover (flip/scale/rotate/filter)
as full thumbs. The layout choice reuses the seeded random_index, so
builds stay reproducible under RANDOM_SEED. THUMB_SUBDIVIDE_PERCENT=0
reproduces the previous output byte-for-byte.
The new option is wired through the config defaults, validation
(0..100), --print-config, the --subdivide CLI flag, the usage text, and
the shuriken.json / --dry-run metadata, with docs and tests updated.
Generated HTML and CSS pass the W3C Nu HTML checker and CSS validator.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
Every shuriken-generated page failed W3C validation. Fixed across the
templates and the bash thumbnail builders so all page categories (splash,
gallery, photo view, details, stats overview, stats filter mini-albums and
redirect stubs) validate cleanly:
- Add <!DOCTYPE html>, <html lang="en"> and <meta charset="utf-8"> to the
header, splash and redirect templates.
- Drop the obsolete type="text/css" on <style> and border='0' on <img>.
- Fix invalid CSS "margin: 2 auto" -> "margin: 2px auto".
- Add required alt attributes to every <img> (splash, thumbnails, views).
- Replace the obsolete name attribute on <a> thumbnail anchors with id (in
album-render.source.sh and stats-filter-album.source.sh, where the markup
is actually built, plus preview.tmpl for consistency).
- Give redirect stubs a <title>; drop trailing slashes on void elements.
Updated the affected cli.sh assertions and rebuilt bin/shuriken.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
The EXIF stats site (stats/ overview plus per-camera and filter mini-albums) was
generated by default. Flip the STATS_PAGE default to "no" so a plain album stays
lean; enable it explicitly with STATS_PAGE=yes or --stats. Updated the bundled
default config, the README ("off by default"), and the tests that relied on the
old default (the stats-rendering and dry-run-override tests now pass --stats /
set STATS_PAGE=yes; the print-config expectations now show STATS_PAGE=no).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
The footer "Site generated ... with <link>" was hardcoded to the shuriken.sh
repository. Make it configurable via the SOURCE_URL config variable and the
--source-url CLI flag (defaulting to the shuriken.sh repo, so existing sites are
unchanged). The footer derives the displayed text from the URL by stripping its
scheme. Plumbed through apply_config_defaults, CLI override targets/spec,
--print-config, the verbose effective-config log, and the header template's
new render_source_url_html (config_html) render var. Documented in
shuriken.default.conf and README; added a generation test asserting a custom
SOURCE_URL replaces the default footer link, and updated the print-config and
header render-var-subset expectations.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
Stats reached directly into album-module internals: the private global
ALBUM_VIEW_PAGE_BY_PHOTO and the EXIF cache reader
cached_photo_identify_output. Introduce a clean boundary, behaviour and
generated HTML byte-identical.
- Promote the EXIF identify cache primitive (cached_photo_identify_output
plus its private helpers photo_cache_signature and
print_cached_photo_identify_output) out of album-metadata.source.sh into
a new shared src/lib/metadata-cache.source.sh, sourced before both album
and stats (right after metadata-label in LIB_SOURCES). It is a low-level
metadata primitive used by both consumers, so it no longer belongs to
album internals. Signature/behaviour unchanged.
- Add album_view_page_for_photo accessor in album-render.source.sh as the
documented public API; keep ALBUM_VIEW_PAGE_BY_PHOTO as the album's
private backing store. stats-filter-album.source.sh now calls the
accessor instead of indexing the global, so a change to album page
naming/caching stays contained in the album module.
- Add test_album_stats_decoupling_boundary asserting the accessor returns
the backing-store value and that the assembled bin/shuriken keeps the
cache primitive in the shared module and no longer indexes the global
from the stats filter section. Existing stats/album tests unchanged.
Verified: 3-image fixture (STATS_PAGE=yes, fixed seed) diff -r of stashed
original vs new build is byte-identical across all 47 dist files (only the
inherent generated_at timestamp normalized). just test, just shellcheck,
just check-generated and git diff --check all pass.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
prepare_template_render_vars previously computed and serialized every one
of the 30+ render_* fields for each template invocation, even though e.g.
the header template references only a handful. It now computes only the
subset the target template actually needs.
The needed set is driven by the required_templates (5th) field of each
TEMPLATE_RENDER_FIELD_SPECS entry, corrected/completed so every spec's
required_templates exactly matches that render_var's references in the
.tmpl files (config_html and derived kinds previously left it empty). A
new template_needed_render_vars_to builds the per-template set;
prepare_template_render_vars takes the template name and skips non-needed
fields; serialize_template_render_context emits only the computed keys
into the BASH_ENV context file.
Dependency closure: handlers read only from the input context array or
config globals, never from another computed render_var, so the direct
per-template set is the full closure (no transitive expansion needed).
Side-effects: every handler is a pure value computation; none consume
RANDOM/seed; current_date_html only primes the deterministic
SHURIKEN_CURRENT_DATE_TEXT cache (idempotent), so subsetting is safe for
all fields. render_html_dir_html (required_templates='*') is referenced by
no template but kept always-computed as a documented cheap exception.
Output is byte-identical: diff -r over a full generated dist/ (3-image,
2-camera fixture, STATS_PAGE=yes, fixed random seed) between the previous
bin/shuriken and this build matches exactly across all 63 files (51 HTML
album/details/splash/stats/per-camera/filter pages + tarball).
Tests: add test_template_render_vars_subset_is_minimal_for_header (header
computes only its needed vars, succeeds with unrelated config globals
unset) and test_template_render_var_subsetting_matches_templates (spec
needed-set equals each .tmpl's render_* references). Existing
test_template_required_context_vars_come_from_render_specs is unchanged.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
Adding a CLI action previously required editing two case statements
(run_action and run_configured_action). Introduce ACTION_SPECS, a single
'|'-delimited registry (flag|handler|requires_config|validation_fn|
validation_arg) matching the CLI_OPTION_SPEC / STATS_CATEGORIES encoding,
and replace both dispatchers with table lookups via action_spec_field.
- run_action: looks up requires_config; routes non-config actions
(--version/--init) through run_unconfigured_action (shared
config/override/force precheck) and the rest through
run_configured_action.
- run_configured_action: keeps the force-generate guard and config
load/log, then runs the entry's validation_fn (with optional arg, used
by --dry-run) and handler via run_configured_action_body.
- Extracted the --clean inline rm body into a clean_dist handler and
added an action_print_version handler so every action is just a
registry entry plus named functions.
- Unknown/empty actions have no entry, so dispatch falls through to the
same usage + exit 1 behavior as the old case "*)" arm.
Adding an action is now one ACTION_SPECS entry plus its handler/
validation functions; neither dispatcher changes (Open/Closed).
Added test_action_dispatch_is_registry_driven proving every parser
action flag has a registry entry, all handlers/validators resolve, and
unknown actions are rejected. All existing action/dispatcher tests pass
unchanged.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
Adding an EXIF stats category previously required editing four places: the
reset function, a new _stats_record_*, the body builder, and a new
_stats_render_* section. Introduce a STATS_CATEGORIES registry (the single
source of truth) and make the generic code iterate it instead.
- STATS_CATEGORIES: ordered, pipe-delimited specs
(count_array|prefix|heading|render_kind), declared -gra so it survives a
function-scoped source. The array order IS the overview display order.
- STATS_CATEGORY_BUCKETS: tab-delimited bucket ladders for the 'ordered'
histogram kinds (apertures wide->narrow, etc.).
- STATS_RECORD_FUNCTIONS: the per-photo recorder dispatch list.
Collapsed touch-points:
- reset_photo_exif_stats clears each registry count array via
_stats_category_arrays.
- accumulate_photo_stats dispatches recorders from STATS_RECORD_FUNCTIONS.
- _stats_build_body iterates STATS_CATEGORIES, dispatching each spec through
_stats_render_category (camera/ranked/ordered/month kinds).
- _stats_render_ordered_section reads its ladder from STATS_CATEGORY_BUCKETS;
the camera leaderboard is now ranked + a 'stats-leaderboard' list_class.
Adding a category is now: append one STATS_CATEGORIES entry (plus a
STATS_CATEGORY_BUCKETS row for an ordered ladder) and have a record function
tally into its array. No edits to reset, the body builder, or a per-category
render branch.
Behaviour-preserving: category order, bucket order, headings, counts and
links are unchanged. Verified byte-identical by diffing the stats/ output of
the pre-change binary against the new one over the same fixture album.
Added test_stats_categories_registry_is_single_source_of_truth (registered in
main()) asserting reset, the body builder and the bucket ladders all derive
from the registry; it fails if a category is added in only one place.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
The --clean action deleted DIST_DIR but left behind the staging/backup
directories the generation pipeline creates as siblings of DIST_DIR
(.shuriken.<basename>.staging.* / .backup.*). Users expect --clean to
remove all generation output, so extend it to also delete those.
Cleanup runs after the validate_clean_dist_dir safety guard (8n0), so a
dangerous DIST_DIR still aborts before any deletion. It only matches
shuriken's own basename-specific staging/backup prefixes (the exact
mktemp templates from config.staging.source.sh), uses nullglob so a
missing match never expands to a literal pattern, and only removes
directories. Unrelated dotfiles in the parent are never touched.
Update test_clean to assert the staging/backup dirs are removed while
unrelated entries survive, and document the behavior in the CLI usage
text and README.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
The rule that joins a camera's EXIF Make + Model into one label while
avoiding a duplicated manufacturer prefix (e.g. "Canon Canon EOS 5D" ->
"Canon EOS 5D") was implemented independently in the album tooltip builder
and the stats leaderboard tally. Extract it into a single shared helper
camera_label_from_make_model in the new src/lib/metadata-label.source.sh,
sourced before both callers.
Both prior implementations were behavior-identical (empty model -> make,
empty make -> model, exact/prefix dedup, case-sensitive), so this is a pure
DRY refactor with no observable output change. Added a focused unit test
covering dedup, plain concatenation and the empty-field edge cases.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
Shuriken is a single-process CLI, yet the action layer could serialize 30+
globals plus every function definition (declare -p / declare -f) and pipe them
into a fresh "bash -euo pipefail" process to run an action. Production already
forced the in-process run_action_body_direct via SHURIKEN_ACTION_BODY_RUNNER, so
the serialized-subprocess path was dead in production and only added complexity
(a hand-maintained variable list to keep in sync). Per KISS, drop it.
- Remove run_action_body_context and the run_action_body dispatcher.
- Collapse run_configured_action_body to call the action in-process directly and
remove the SHURIKEN_ACTION_BODY_RUNNER indirection in main().
- Move the only genuinely needed isolation into a test-only shim
(test::run_action_isolated in tests/helpers.sh) for the generate real-failure
test, which must capture a failure status without the in-process errexit abort
ending the caller (correct in production, where main runs under errexit).
- Update the errexit/status-propagation tests to exercise the direct path.
Template-engine serialization is unrelated and left untouched.
just test, just shellcheck, just check-generated and git diff --check all pass.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
A scalar SYNC_DESTINATIONS containing spaces was word-split by the shared
resolve_config_array helper, breaking a single destination into multiple
broken arguments passed to rsync.
A list of rsync destinations is inherently a list, and array syntax is the
only spelling that preserves embedded spaces. resolve_sync_destinations now
detects a scalar declaration via declare -p and fails with a clear
config_error telling the user to use array syntax. The array path is
unchanged, and resolve_config_array's scalar word-splitting is left intact
for TAR_OPTS (where turning "-c -v" into separate options is desired).
Adds focused tests proving the scalar case errors without invoking rsync and
the array case preserves a space-containing destination as one argument.
Updates src/shuriken.default.conf and README.md to document the requirement.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
The preview_num next/prev render handlers computed neighbour page numbers
with $(( context_value +/- 1 )) but only guarded against an empty value.
A non-numeric preview_num context value (e.g. a stray string) slipped past
the [ -n ] check and triggered a bash arithmetic syntax error which, under
set -e, aborted the whole script.
Validate the context value is a non-negative integer
([[ value =~ ^[0-9]+$ ]]) before the arithmetic in both
prepare_template_render_var__preview_num_next_html and __preview_num_prev_html.
Invalid or missing values now default to an empty render value, matching the
existing missing-neighbour behaviour; the valid-numeric path is unchanged.
Add test_template_render_var_preview_num_guards_non_numeric in tests/cli.sh,
which drives the handlers directly under bash -euo pipefail to prove a bad
preview_num no longer crashes and a numeric one still yields the exact
+1 / -1 neighbour.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
The --clean action ran `rm -rf "$DIST_DIR"` after only an `[ -d ]` check,
so a misconfigured DIST_DIR (empty, /, $HOME, system dirs, etc.) could
recursively delete the wrong tree.
Add validate_clean_dist_dir (and resolve_dist_dir_path) in
config.validate.source.sh and call it in the --clean case before any
deletion. The guard canonicalizes DIST_DIR with `pwd -P` (handling ./
trailing slashes, symlinks and relative paths; for a not-yet-existing
dir it resolves the existing parent and re-attaches the basename) and
refuses to clean when the resolved path is empty, the filesystem root, a
well-known system directory, the resolved $HOME, or the current working
directory. Rejection uses config_error with a clear message and a
non-zero exit, so nothing is deleted. Normal DIST_DIRs still clean.
Tests (tests/cli.sh, registered in main): a HOME-as-DIST_DIR case (uses
a fake HOME under TEST_TMPDIR with a sentinel file, so a regression can
only touch the throwaway temp dir) and an empty-DIST_DIR case both assert
rejection and that nothing is removed.
Note: leftover staging artifacts on --clean are out of scope (task ln0).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
cached_photo_identify_output() swallowed all ImageMagick errors with
'imagemagick_identify ... || true', so a corrupt photo or identify
failure left the cache with only the signature line and no EXIF. That
signature-only file was a valid-looking cache hit, so the photo rendered
with empty tooltip/stats and the failure was never retried or warned
about again.
Now capture identify's exit status; on failure, log_warning naming the
photo and rm -f the cache file so the next run retries instead of
reusing an empty result. The function still returns 0 so one unreadable
photo does not abort generation (it runs in backgrounded render jobs
under set -euo pipefail) -- the photo just renders without EXIF, now
with a warning.
Adds tests/cli.sh test_generate_warns_and_skips_cache_on_identify_failure
and a TEST_IMAGEMAGICK_IDENTIFY_FAIL hook in the fake ImageMagick to
drive a failing identify; asserts exit 0, a warning naming the photo,
the photo still rendered, and that the failed photo's cache is absent
while a successful photo's cache is present.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
Each thumbnail on a preview page (page-N.html) used to be rendered by its
own "template preview" call, paying the full source_template_file cost --
an "env -i bash" invocation -- per thumbnail. With MAXPREVIEWS thumbnails
per page that was N template renders per page just for the grid.
render_full_preview_page now builds the markup for ALL of a page's
thumbnails in bash (build_preview_thumbnail / append_preview_thumbnail)
and emits the whole grid in ONE render via a new previewpage.tmpl that
takes the pre-built HTML through a context_raw "preview_thumbs" field --
the same pattern the stats filter galleries (camera.tmpl) already use.
Per-thumbnail markup is byte-identical to the old preview.tmpl output:
same <a name=... href=...><img class='thumb <anim>' .../></a> structure,
order, HTML escaping and seeded "slow" animation class. Header and footer
stay as their own template calls, so a page now costs ~1 previewpage
render + header/footer instead of N + chrome.
The parallel job-pool integration and failure contract are unchanged:
each preview page is still one background render job. Added the
render_preview_thumbs_html field spec (hn0 dispatch pattern, context_raw
kind), registered previewpage in the validate_template_dir required
templates and in the required-context-vars test expectations, and pointed
the four generation template-failure tests at previewpage.tmpl (generation
no longer renders preview.tmpl). The standalone "template preview" engine
unit tests keep exercising preview.tmpl, which still ships.
Verified byte-identical output: generated the fixture album (including a
spaces/special-char filename) twice with RANDOM_SEED=42 using the parent
commit's bin/shuriken vs the new bin; every .html file is identical.
just test, just shellcheck, just check-generated and git diff --check all
pass.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
Replace the hardcoded `case "$kind"` block in
prepare_template_render_vars with a name-based registration/dispatch
pattern. Each render field kind is now implemented by one
prepare_template_render_var__<kind> handler; the core loop resolves the
handler by name (prepare_template_render_var__$kind), verifies it exists
via `declare -F`, calls it with a uniform signature
(out_nameref, context_name, source_name), and reports a config_error for
an unknown kind (no matching handler) -- preserving the previous error
behavior. Adding a new kind now means defining a new handler function
only; the loop never changes.
Handlers cover all existing kinds: context_css, context_html,
context_raw, current_date_html, config_html (keeps its inner source_name
dispatch for HEIGHT/MAXPREVIEWS/TITLE/etc. and the same :- defaults),
original_basepath_is_set, preview_num_next_html, preview_num_prev_html,
and tarball_include. Escaping and defaults are unchanged, so rendered
output is byte-identical (verified by diff -r of a full --generate
album, before vs after).
Add test_template_render_var_dispatch_is_extensible proving every
declared kind resolves to a handler and that a newly defined handler is
dispatched without touching the core loop (OCP). just test, just
shellcheck, just check-generated and git diff --check all pass.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
source_template_file previously built the BASH_ENV context file by piping
"declare -p ...; declare -f; serialize_template_render_context ...;" into a
fresh "bash -euo pipefail" subprocess for every rendered page. That dumped all
~5000 lines of shuriken functions and spawned a subprocess per page just to run
the serializer - a large per-page cost for albums with hundreds-to-thousands of
pages.
Now serialize_template_render_context runs in the current shell with stdout
redirected straight into the context tempfile, followed by an appended
"unset BASH_ENV". The serializer returns its own non-zero status explicitly so
the failure is detected via an "if" status-test (which returns normally through
the RETURN trap and cleans up the partial context file), robust even when
source_template_file runs inside a status-tested "if template ..." call chain
where bash would otherwise suppress an inner errexit abort.
The trap-based cleanup (RETURN plus INT/TERM/HUP re-raising to $BASHPID) is
preserved unchanged. Two serializer test mocks that relied on the removed
"| bash" errexit now return non-zero explicitly. Rendered HTML output is
unchanged; just test, just shellcheck, just check-generated and
git diff --check all pass.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
The generated pages link a favicon.ico that was always the bundled shuriken
favicon (copy_site_favicon hard-copied share/shuriken/assets/favicon.ico). Add a
FAVICON config variable and a --favicon PATH CLI flag: when set, that file is
published as favicon.ico instead of the bundled default; when empty, the bundled
favicon is used as before.
Plumbed through apply_config_defaults, CLI_OPTION_SPEC + override allowlist,
usage, print_config, the action config list and effective-setting log, and
validated (a non-empty FAVICON must be a readable file) before generation.
shuriken.default.conf and the README document it; a test covers a custom favicon,
its appearance in --print-config, and rejection of a missing file.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
The dist directory is the swapped-in staging dir, which mktemp -d creates mode
0700. The published album root therefore stayed 0700, so the first `shuriken
--sync` created the remote album directory 0700 and the web server (daemon) could
not read it -- requiring a manual chmod 755 on each mirror.
Relax the staging dir to the umask-default directory mode right after mktemp -d
(what mkdir would have produced), so the dist root matches its subdirectories and
is served/synced with sane permissions. Add a regression test asserting the dist
root mode equals its photos/ subdirectory mode.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
|
The filter mini-albums put thousands of HTML files directly in the album root.
Reorganise so only the main album lives in DIST_DIR and all stats content goes
under stats/:
stats/index.html - the stats overview (was stats.html)
stats/<pagebase>/index.html - each filter gallery (was <pagebase>.html)
stats/<pagebase>/<index>.html - each fi |
