diff options
Diffstat (limited to 'wqflask')
-rw-r--r-- | wqflask/wqflask/metadata_edits.py | 163 |
1 files changed, 81 insertions, 82 deletions
diff --git a/wqflask/wqflask/metadata_edits.py b/wqflask/wqflask/metadata_edits.py index 7c701008..49205b76 100644 --- a/wqflask/wqflask/metadata_edits.py +++ b/wqflask/wqflask/metadata_edits.py @@ -19,6 +19,7 @@ from flask import render_template from flask import request from flask import url_for +from wqflask.database import database_connection from wqflask.decorators import edit_access_required from wqflask.decorators import edit_admins_access_required from wqflask.decorators import login_required @@ -210,10 +211,6 @@ def display_probeset_metadata(name: str): @edit_access_required @login_required def update_phenotype(dataset_id: str, name: str): - conn = MySQLdb.Connect(db=current_app.config.get("DB_NAME"), - user=current_app.config.get("DB_USER"), - passwd=current_app.config.get("DB_PASS"), - host=current_app.config.get("DB_HOST")) data_ = request.form.to_dict() TMPDIR = current_app.config.get("TMPDIR") author = ((g.user_session.record.get(b"user_id") or b"").decode("utf-8") @@ -231,18 +228,20 @@ def update_phenotype(dataset_id: str, name: str): current_time = str(datetime.datetime.now().isoformat()) _file_name = (f"{author}.{request.args.get('resource-id')}." f"{current_time}") - diff_data = remove_insignificant_edits( - diff_data=csv_diff( - base_csv=(base_csv:=get_trait_csv_sample_data( - conn=conn, - trait_name=str(name), - phenotype_id=str(phenotype_id))), - delta_csv=(delta_csv:=file_.read().decode()), - tmp_dir=TMPDIR), - epsilon=0.001) + diff_data = {} + with database_connection() as conn: + diff_data = remove_insignificant_edits( + diff_data=csv_diff( + base_csv=(base_csv := get_trait_csv_sample_data( + conn=conn, + trait_name=str(name), + phenotype_id=str(phenotype_id))), + delta_csv=(delta_csv := file_.read().decode()), + tmp_dir=TMPDIR), + epsilon=0.001) # Edge case where the csv file has not been edited! if not any(diff_data.values()): - flash(f"You have not modified the csv file you downloaded!", + flash("You have not modified the csv file you downloaded!", "warning") return redirect(f"/datasets/{dataset_id}/traits/{name}" f"?resource-id={request.args.get('resource-id')}") @@ -284,10 +283,12 @@ def update_phenotype(dataset_id: str, name: str): "owner": data_.get("owner"), "authorized_users": data_.get("authorized-users"), } - updated_phenotypes = update( - conn, "Phenotype", - data=Phenotype(**phenotype_), - where=Phenotype(id_=data_.get("phenotype-id"))) + updated_phenotypes = "" + with database_connection() as conn: + updated_phenotypes = update( + conn, "Phenotype", + data=Phenotype(**phenotype_), + where=Phenotype(id_=data_.get("phenotype-id"))) diff_data = {} if updated_phenotypes: diff_data.update({"Phenotype": diff_from_dict(old={ @@ -303,11 +304,13 @@ def update_phenotype(dataset_id: str, name: str): "month": data_.get("month"), "year": data_.get("year") } - updated_publications = update( - conn, "Publication", - data=Publication(**publication_), - where=Publication(id_=data_.get("pubmed-id", - data_.get("old_id_")))) + updated_publications = "" + with database_connection() as conn: + updated_publications = update( + conn, "Publication", + data=Publication(**publication_), + where=Publication(id_=data_.get("pubmed-id", + data_.get("old_id_")))) if updated_publications: diff_data.update({"Publication": diff_from_dict(old={ k: data_.get(f"old_{k}") for k, v in publication_.items() @@ -323,11 +326,12 @@ def update_phenotype(dataset_id: str, name: str): .now() .strftime("%Y-%m-%d %H:%M:%S")), }) - insert(conn, - table="metadata_audit", - data=MetadataAudit(dataset_id=name, - editor=author, - json_data=json.dumps(diff_data))) + with database_connection() as conn: + insert(conn, + table="metadata_audit", + data=MetadataAudit(dataset_id=name, + editor=author, + json_data=json.dumps(diff_data))) flash(f"Diff-data: \n{diff_data}\nhas been uploaded", "success") return redirect(f"/datasets/{dataset_id}/traits/{name}" f"?resource-id={request.args.get('resource-id')}") @@ -365,12 +369,12 @@ def update_probeset(name: str): "flag": data_.get("flag") } diff_data = {} - author = ((g.user_session.record.get(b"user_id") or b"").decode("utf-8") + author = ((g.user_session.record.get(b"user_id") + or b"").decode("utf-8") or g.user_session.record.get("user_id") or "") - if (updated_probeset := update( - conn, "ProbeSet", - data=Probeset(**probeset_), - where=Probeset(id_=data_.get("id")))): + if update(conn, "ProbeSet", + data=Probeset(**probeset_), + where=Probeset(id_=data_.get("id"))): diff_data.update({"Probeset": diff_from_dict(old={ k: data_.get(f"old_{k}") for k, v in probeset_.items() if v is not None}, new=probeset_)}) @@ -391,8 +395,8 @@ def update_probeset(name: str): @metadata_edit.route("/<dataset_id>/traits/<phenotype_id>/csv") @login_required -def get_sample_data_as_csv(dataset_id: str, phenotype_id: int): - with database_connection as conn: +def get_sample_data_as_csv(dataset_id: str, phenotype_id: int): + with database_connection() as conn: return Response( get_trait_csv_sample_data( conn=conn, @@ -459,70 +463,67 @@ def reject_data(resource_id: str, file_name: str): @metadata_edit.route("<resource_id>/diffs/<file_name>/approve") @edit_admins_access_required @login_required -def approve_data(resource_id:str, file_name: str): +def approve_data(resource_id: str, file_name: str): sample_data = {file_name: str} - conn = MySQLdb.Connect(db=current_app.config.get("DB_NAME"), - user=current_app.config.get("DB_USER"), - passwd=current_app.config.get("DB_PASS"), - host=current_app.config.get("DB_HOST")) TMPDIR = current_app.config.get("TMPDIR") with open(os.path.join(f"{TMPDIR}/sample-data/diffs", file_name), 'r') as myfile: sample_data = json.load(myfile) - for modification in ( - modifications := [d for d in sample_data.get("Modifications")]): - if modification.get("Current"): - update_sample_data( - conn=conn, - trait_name=sample_data.get("trait_name"), - original_data=modification.get("Original"), - updated_data=modification.get("Current"), - csv_header=sample_data.get("Columns", - "Strain Name,Value,SE,Count"), - phenotype_id=int(sample_data.get("phenotype_id"))) + with database_connection() as conn: + for modification in ( + modifications := [d for d in + sample_data.get("Modifications")]): + if modification.get("Current"): + update_sample_data( + conn=conn, + trait_name=sample_data.get("trait_name"), + original_data=modification.get("Original"), + updated_data=modification.get("Current"), + csv_header=sample_data.get("Columns", + "Strain Name,Value,SE,Count"), + phenotype_id=int(sample_data.get("phenotype_id"))) n_deletions = 0 - for data in [d for d in sample_data.get("Deletions")]: - __deletions = delete_sample_data( - conn=conn, - trait_name=sample_data.get("trait_name"), - data=data, - csv_header=sample_data.get("Columns", - "Strain Name,Value,SE,Count"), - phenotype_id=int(sample_data.get("phenotype_id"))) - if __deletions: - n_deletions += 1 - # Remove any data that already exists from sample_data deletes - else: - sample_data.get("Deletions").remove(data) - - n_insertions = 0 - for data in [d for d in sample_data.get("Additions")]: - if insert_sample_data( + with database_connection() as conn: + for data in [d for d in sample_data.get("Deletions")]: + __deletions = delete_sample_data( conn=conn, trait_name=sample_data.get("trait_name"), data=data, csv_header=sample_data.get("Columns", "Strain Name,Value,SE,Count"), - phenotype_id=int(sample_data.get("phenotype_id"))): - n_insertions += 1 - # Remove any data that already exists from sample_data inserts - else: - sample_data.get("Additions").remove(insertion) + phenotype_id=int(sample_data.get("phenotype_id"))) + if __deletions: + n_deletions += 1 + # Remove any data that already exists from sample_data deletes + else: + sample_data.get("Deletions").remove(data) + + n_insertions = 0 + with database_connection() as conn: + for data in [d for d in sample_data.get("Additions")]: + if insert_sample_data( + conn=conn, + trait_name=sample_data.get("trait_name"), + data=data, + csv_header=sample_data.get("Columns", + "Strain Name,Value,SE,Count"), + phenotype_id=int(sample_data.get("phenotype_id"))): + n_insertions += 1 if any([sample_data.get("Additions"), sample_data.get("Modifications"), sample_data.get("Deletions")]): - insert(conn, - table="metadata_audit", - data=MetadataAudit( - dataset_id=sample_data.get("trait_name"), - editor=sample_data.get("author"), - json_data=json.dumps(sample_data))) + with database_connection() as conn: + insert(conn, + table="metadata_audit", + data=MetadataAudit( + dataset_id=sample_data.get("trait_name"), + editor=sample_data.get("author"), + json_data=json.dumps(sample_data))) # Once data is approved, rename it! os.rename(os.path.join(f"{TMPDIR}/sample-data/diffs", file_name), os.path.join(f"{TMPDIR}/sample-data/diffs", f"{file_name}.approved")) - message = "" if n_deletions: flash(f"# Deletions: {n_deletions}", "success") if n_insertions: @@ -535,6 +536,4 @@ def approve_data(resource_id:str, file_name: str): f"{file_name}.rejected")) flash(("Automatically rejecting this file since no " "changes could be applied."), "warning") - conn.close() return redirect(url_for('metadata_edit.list_diffs')) - |