CLI: add stdin support for instance validation by Vaibhav701161 · Pull Request #614 · sourcemeta/jsonschema

Vaibhav701161 · 2026-01-15T09:35:35Z

Adds stdin support for instance input to enable real-time editor diagnostics.
Fixes(#537)

Centralized stdin parsing in src/input.h (JSON-first, YAML fallback).
validate accepts instances from stdin only; schemas must be file paths.
Stdin inputs resolve relative $ref against the current working directory.
Includes integration test test/validate/pass_stdin.sh.
Unblocks sourcemeta/studio#61; addresses sourcemeta/jsonschema#537.

Refs: sourcemeta/studio#61, #537

*Build and test:

Vaibhav701161 · 2026-01-17T05:37:30Z

Hi @jviotti , I’ve opened this PR and would appreciate a review when you have time. Thanks!

jviotti · 2026-01-19T12:37:22Z

Hey @Vaibhav701161 , sorry for the delay. It has been a hectic week. I'll try to get into it today

jviotti · 2026-01-19T12:54:42Z

I like the idea but I think there are some things to iron out:

If we accept standard input, we should do so in ALL commands, otherwise it gets weird
Some commands require a schema as input but for others, passing nothing means "the current directory", which introduces some ambiguity. For example, jsonschema fmt formats all schemas in the current directory, while jsonschema inspect <schema.json> requires a single schema argument
Furthermore, some commands require 2 schemas, like validate

I think for consistency, we can require the - to signify standard input in every command. However, for commands take potentially take directories as input, we should transparently implement this logic on src/input.h. So it happens out of the box.

The other problem here will be line numbers. We should make sure all relevant commands handle standard input while still reporting line/column information and potentially some "special" file path like <stdin>?

Vaibhav701161 · 2026-01-20T04:03:35Z

Hey @jviotti , thanks a lot for the feedback.

I’ve pushed an updated iteration that tries to address the concerns you mentioned:

- is now the explicit opt-in for stdin.
The stdin logic is centralized in src/input.h, so it’s not handled ad-hoc in individual commands.
Ambiguous cases (schema via stdin, multiple stdin inputs) are explicitly rejected.
Line/column information is preserved, and stdin is handled as a special input (covered via tests).

Signed-off-by: Vaibhav mittal <vaibhavmittal929@gmail.com>

Vaibhav701161 · 2026-02-01T13:45:46Z

Hi @jviotti , just a quick follow-up on this.
Whenever you get a chance, I’d appreciate a review to confirm whether this direction looks good or if there’s anything you’d like adjusted before moving forward. Thanks

jviotti · 2026-01-21T13:18:19Z

src/input.h

+inline auto read_from_stdin() -> ParsedJSON {
+  std::string buffer;
+  {
+    std::ostringstream ss;


You shouldn't need this. sourcemeta::core::parse_json should already be able to read from any standard C++ input stream without needing to buffer it

This too is still valid

jviotti · 2026-01-21T13:22:33Z

src/input.h


 struct InputJSON {
  std::filesystem::path first;
+  std::filesystem::path resolution_base;


If this is just a display name, maybe make a std::string? Otherwise <stdin> is not really a path

I think this still needs to be addressed?

jviotti · 2026-01-21T13:23:20Z

test/validate/pass_stdin.sh

Please split this test into multiple tests. If you see the other tests in the project, we have specific conventions and we always test a single thing per file. Plus you also want to add test for every other applicable command.

Finally, you didn't register this test in test/CMakeLists.txt, so it won't run at all

jviotti · 2026-02-05T18:31:10Z

src/command_lint.cc

                  copy, sourcemeta::core::schema_walker, custom_resolver,
                  get_lint_callback(errors_array, entry, output_json), dialect,
-                  sourcemeta::jsonschema::default_id(entry.first),
+                  sourcemeta::core::URI::from_path(entry.resolution_base)


I think we still definitely want to use sourcemeta::jsonschema::default_id here, to concentrate that logic in a single place. Maybe make such function take entry directly, so that inside we can decide to use entry.resolution_base or whatever?

jviotti · 2026-02-05T18:31:16Z

src/command_lint.cc

                  custom_resolver,
                  get_lint_callback(errors_array, entry, output_json), dialect,
-                  sourcemeta::jsonschema::default_id(entry.first),
+                  sourcemeta::core::URI::from_path(entry.resolution_base)


jviotti · 2026-02-05T18:33:27Z

src/input.h

        // TODO: Print a verbose message for what is getting parsed
        auto parsed{read_file(canonical)};
-        result.push_back({std::move(canonical), std::move(parsed.document),
+        auto canonical_copy{canonical};


jviotti · 2026-02-05T18:33:37Z

src/input.h

    }
  }
 }
-


I don't think this deletion was intended?

jviotti · 2026-02-05T18:34:43Z

src/input.h

+    }
+  }
+  if (stdin_count > 1) {
+    throw std::runtime_error{"Standard input (-) can only be specified once"};


Suggested change

throw std::runtime_error{"Standard input (-) can only be specified once"};

throw std::runtime_error{"The standard input position argument `-` can only be specified once"};

jviotti · 2026-02-05T18:35:17Z

src/input.h

+  std::size_t stdin_count{0};
+  for (const auto &arg : arguments) {
+    if (arg == "-") {
+      stdin_count++;


I don't think counting all occurrences makes sense if you only want up to two. Maybe check that if you already got at least 2, then its an error and you can break?

jviotti · 2026-02-05T18:36:57Z

test/validate/pass_stdin.sh

+set +o errexit
+cat "$TMP/schema.json" > "$TMP/schema_stdin.json"
+
+OUTPUT=$("$1" validate - "$TMP/schema.json" < "$TMP/schema_stdin.json" 2>&1)


Please see the other shell scripts for the conventions on how we test stuff. Some obvious ones:

We never capture into a variable. We pipe into a file and then diff it

We never enable set +o errexit

We never test more than one thing at once

Also your test is not registered in test/CMakeLists.txt so it won't run at all. But overall, split it so that there are multiple tests testing each one thing

Vaibhav701161 force-pushed the feat/stdin-validate branch 2 times, most recently from ddb8312 to 9a040d7 Compare January 15, 2026 10:07

Vaibhav701161 force-pushed the feat/stdin-validate branch from 4d5ff3f to 112b374 Compare January 20, 2026 02:49

Vaibhav701161 added 7 commits February 1, 2026 12:24

Feat: support reading instances from stdin for validate and lint

9ae956a

Signed-off-by: Vaibhav mittal <vaibhavmittal929@gmail.com>

test: fix shellcheck SC2002 in stdin validation test

efec465

Signed-off-by: Vaibhav mittal <vaibhavmittal929@gmail.com>

test: fix shellcheck SC2002 in stdin validation test-2

51974d3

Signed-off-by: Vaibhav mittal <vaibhavmittal929@gmail.com>

feat:centralize stdin handling and add resolution context

28be6af

Signed-off-by: Vaibhav mittal <vaibhavmittal929@gmail.com>

style: apply clang-format

cf8a872

Signed-off-by: Vaibhav mittal <vaibhavmittal929@gmail.com>

test: cover stdin edge cases and fix shellcheck warnings

dead00c

Signed-off-by: Vaibhav mittal <vaibhavmittal929@gmail.com>

test: cover stdin edge cases and fix shellcheck warnings-2

b9bb23e

Signed-off-by: Vaibhav mittal <vaibhavmittal929@gmail.com>

Vaibhav701161 force-pushed the feat/stdin-validate branch from f70d55d to b9bb23e Compare February 1, 2026 13:22

jviotti requested changes Feb 5, 2026

View reviewed changes

	throw std::runtime_error{"Standard input (-) can only be specified once"};
	throw std::runtime_error{"The standard input position argument `-` can only be specified once"};

Conversation

Vaibhav701161 commented Jan 15, 2026

Uh oh!

Vaibhav701161 commented Jan 17, 2026

Uh oh!

jviotti commented Jan 19, 2026

Uh oh!

jviotti commented Jan 19, 2026

Uh oh!

Vaibhav701161 commented Jan 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Vaibhav701161 commented Feb 1, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Vaibhav701161 commented Jan 20, 2026 •

edited

Loading